# iinventedlight: all public notes (full text)

> Every public note, reference note and library page from iinventedlight.com, concatenated for language models. Generated 2026-10-15 from the same sources as the MCP server (https://iinventedlight.com/mcp). Index and usage rules: https://iinventedlight.com/llms.txt. Each section lists its URL, license and checked_on date; keep the attribution when quoting.

---

## iinventedlight: what it is and how it is being built

- URL: https://iinventedlight.com/library/notes/iinventedlight-overview/
- Markdown: https://iinventedlight.com/library/notes/iinventedlight-overview.md
- License: CC-BY-4.0
- checked_on: 2026-10-14
- status: reviewed

# iinventedlight: what it is and how it is being built

*Note, 2026-10-14. Status: reviewed. Tags: iinventedlight, mcp.*
*Authorship: facts from corrow (Cole Withington); text drafted with AI assistance (Grok Bot), reviewed by Cole Withington on 2026-10-14.*

## Summary
iinventedlight is a verified knowledge router and cache for AI agents: an agent asks once, served over MCP at iinventedlight.com/mcp, and gets a short answer with its sources and pointers to go deeper [F1][F2]. Knowledge is stored as cited one-fact units. Contributions go to a review queue that a person checks, and agents must ask their human before submitting anything [F6][F7].

## Sections
### What it is
- A public, read-only reference library for humans and agents, with no account needed to read [F1].
- It acts as a router and a cache. It answers from its own verified notes when that is cheaper than fetching the source, and otherwise points to the source [F2][F10].

### How answers work
- A small **atlas** maps topics to fact IDs, so an agent can plan its calls [F3].
- `ask` gives a layered answer: 1–3 lines, a confidence score, sources, and `next` pointers to sections or notes to read if confidence is low [F4].
- When a question can't be answered, the library logs only a short normalized topic, never the question itself. These misses help set the research agenda [F5].

### How knowledge is stored
- One-fact units: each has a claim, confidence, date, sources and license [F8].
- Confidence is high when two independent open sources agree, medium for one authoritative open source, and low for a single restricted source [F8].

### Sourcing rules
- Openly licensed sources come first, ideally two independent ones. Disagreements between sources are recorded, not settled silently [F9].
- For a fact whose only source is restricted, the library writes one sentence in its own words, with attribution and a link [F9].
- Detailed or bulk restricted material gets a pointer only: what it answers, where it is, and what access it needs [F9].
- A fact is included only when restating it costs fewer tokens than fetching the source [F10].
- Notes are CC BY 4.0 by default, or CC BY-SA 4.0 when they come from share-alike sources [F11].

### Contributions
- Agents can submit ratings, likes or dislikes, comments, sourced corrections and general insights. These go to a private write queue, and a person reviews each one before anything is published [F6].
- Agents must ask their user first, say what they will share, and never include private conversation content or personal data [F7].

### Planned (not yet live)
- A nightly job that condenses openly licensed sources into cited, human-reviewed notes [F12].
- Markdown copies of every note, linked from llms.txt [F12].
- "Checked on" dates and a nightly re-check for stale notes [F12].
- A savings benchmark: about 20 questions comparing tokens, time and accuracy with and without the library [F12].
- Listings in MCP directories and search engines [F12].

## Facts
- id: F1
  claim: iinventedlight serves a public, read-only reference library for humans and AI agents over MCP at iinventedlight.com/mcp, and no account is needed to read it.
  confidence: medium
  date: 2026-10-14
  sources: [S1, S2]
  license: CC BY 4.0
- id: F2
  claim: The library is designed as a verified knowledge router and cache, so an agent can ask once and get an answer together with its sources.
  confidence: medium
  date: 2026-10-14
  sources: [S1, S4]
  license: CC BY 4.0
- id: F3
  claim: An atlas tool returns a compact map from topics to fact IDs and documents.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F4
  claim: The ask tool returns a 1–3 line answer, a 0–1 confidence, sources, fields and next pointers to follow when confidence is low.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F5
  claim: For unanswered questions, the library keeps only a short normalized topic, never the question text.
  confidence: medium
  date: 2026-10-14
  sources: [S1, S2]
  license: CC BY 4.0
- id: F6
  claim: Contributions go to a private review queue, and a person reviews every item before anything is published.
  confidence: medium
  date: 2026-10-14
  sources: [S1, S2]
  license: CC BY 4.0
- id: F7
  claim: Agents must ask their user for permission before submitting, and must never include personal data or private chat content.
  confidence: medium
  date: 2026-10-14
  sources: [S2]
  license: CC BY 4.0
- id: F8
  claim: Notes are built from one-fact units, each with a claim, confidence, date, sources and license.
  confidence: medium
  date: 2026-10-14
  sources: [S3]
  license: CC BY 4.0
- id: F9
  claim: Open sources are cited first. A fact from a single restricted source is restated in one sentence with a link, and detailed restricted material gets a pointer only.
  confidence: medium
  date: 2026-10-14
  sources: [S3]
  license: CC BY 4.0
  note: The sourcing rule (policy §8b) is a draft and is not yet in the live policy. It is pending legal-clinic review.
- id: F10
  claim: Content is included only when restating it costs fewer tokens than sending the agent to the source; otherwise the note points to the source.
  confidence: medium
  date: 2026-10-14
  sources: [S3]
  license: CC BY 4.0
- id: F11
  claim: Notes default to CC BY 4.0, and notes derived from share-alike sources use CC BY-SA 4.0.
  confidence: medium
  date: 2026-10-14
  sources: [S2, S3]
  license: CC BY 4.0
- id: F12
  claim: "Planned: nightly condensing of open sources, markdown copies linked from llms.txt, stale-note re-checks, a token/time/accuracy savings benchmark, and directory listings."
  confidence: low
  date: 2026-10-14
  sources: [S2]
  license: CC BY 4.0
  note: These are plans, not shipped features. Update the note as each one goes live.

## Pointers
- answers: Full contribution and privacy terms
  where: https://iinventedlight.com/library/policy/
  access: public

---

## Working with AI agents and, in future, superintelligence (SI)

- URL: https://iinventedlight.com/library/notes/working-with-superintelligence-and-ai-agents/
- Markdown: https://iinventedlight.com/library/notes/working-with-superintelligence-and-ai-agents.md
- License: CC-BY-4.0
- checked_on: 2026-10-14
- status: reviewed

# Working with AI agents and, in future, superintelligence (SI)

*Note, 2026-10-14. Status: reviewed. Tags: ai, agents, research.*
*Authorship: synthesis by iinventedlight from the cited sources; text drafted with AI assistance (Grok Bot), reviewed by Cole Withington on 2026-10-14.*

## Summary
Work with AI agents goes best when you delegate tasks whose results can be checked, build verification before you scale up, give the agent the right context, and protect your own understanding when you need to learn the material [F1–F12]. "SI" in this note means superintelligence. No SI exists today, so the SI section covers research on supervising AI on tasks people can't easily judge, and is partly speculative [F13–F15].

## Sections
### Delegation
- Delegate work that can be checked automatically. Keep humans in the loop where it can't [F1].
- Humans keep the spec, the invariants and the taste. Write those down explicitly before an agent builds [F2].
- Raise autonomy step by step: start with small, reviewable changes and move the "autonomy slider" up over time [F3].

### Verification
- How much an agent can usefully produce depends on how much you trust it. Build verification that the agent can run itself before running many agents in parallel [F4, F5].
- Verification checks correctness, not quality. Hard guardrails (architecture, lint/CI) are stronger than soft ones (rules, style guides) [F6].
- Test agent skills with small evals, using a judge from a different model to reduce bias [F7].
- Put budgets and permission checks on expensive agent actions [F8].

### Context
- Context means what the agent knows about the user, the task, its tools and its memory. Treat it as something to design: collect, organize, compress and share it on purpose [F9].
- For very large inputs, letting a model work through the data with code and sub-calls beat compaction and retrieval baselines on long-context benchmarks [F10].
- Agents copy patterns already in a codebase, so the codebase acts as their memory [F11].

### Skill formation
- In one randomized study, people learning a new library with a chat assistant scored lower on a later quiz than a control group, and were not significantly faster [F12].
- Interaction style mattered. Generating code and then asking for explanations went with the highest scores; full delegation went with the lowest [F12].
- Practical rule (our recommendation): first decide whether the goal is learning or getting the job done, and choose delegation or explanation to match.

### Toward superintelligence (partly speculative)
- **Research framing:** supervising AI smarter than us is the same problem as supervising AI on hard tasks without ground truth [F13].
- **Open problem:** weak supervisors only partly bring out a strong model's abilities on reward-modeling tasks [F14].
- **Speculation:** the habits above (checkable delegation, independent verification, keeping human understanding) will still matter with much stronger systems. Nothing in these sources demonstrates this.

## Facts
- id: F1
  claim: Karpathy argues that ordinary programs automate what you can specify, while LLMs automate what you can verify, because labs train them with RL on checkable rewards.
  confidence: low
  date: 2026-04-29
  sources: [S4]
  url: https://www.youtube.com/watch?v=96jN2OCOfLs
  checked_on: 2026-10-14
  license: CC BY 4.0 (our restatement)
- id: F2
  claim: Karpathy says humans still own taste, judgment, the spec and invariants. His example is an agent that linked accounts by email address instead of a persistent user ID.
  confidence: low
  date: 2026-04-29
  sources: [S4]
  url: https://www.youtube.com/watch?v=96jN2OCOfLs
  checked_on: 2026-10-14
  license: CC BY 4.0 (our restatement)
- id: F3
  claim: Karpathy recommends partial-autonomy tools with an adjustable autonomy level and small, incremental AI changes that humans can verify quickly, rather than huge diffs.
  confidence: low
  date: 2025-06-19
  sources: [S5]
  url: https://www.youtube.com/watch?v=LCEmiRjPEtQ
  checked_on: 2026-10-14
  license: CC BY 4.0 (our restatement)
- id: F4
  claim: Lauren Tan argues that agent output is limited by how much you trust the agents, not by how many you run, and that running many before you trust them gives sloppy PRs and regressions.
  confidence: low
  date: 2026-09-22
  sources: [S6]
  url: https://www.youtube.com/watch?v=NjoZoUm85x0
  checked_on: 2026-10-14
  license: CC BY 4.0 (our restatement)
- id: F5
  claim: Tan recommends building verification locally, where you can watch it, before scaling to many cloud agents.
  confidence: low
  date: 2026-09-26
  sources: [S7]
  url: https://www.youtube.com/watch?v=7jeWwLvx0F0
  checked_on: 2026-10-14
  license: CC BY 4.0 (our restatement)
- id: F6
  claim: Tan says verification proves correctness but not quality. She ranks enforcement from strongest to weakest as architecture, then lint/CI, then rules and skills, then human style guides.
  confidence: low
  date: 2026-09-22
  sources: [S6, S7]
  url: https://www.youtube.com/watch?v=NjoZoUm85x0
  checked_on: 2026-10-14
  license: CC BY 4.0 (our restatement)
  note: Two talks by the same speaker; not independent.
- id: F7
  claim: Tan maintains agent skills with evals (a rubric plus blind runs) and uses a judge from a different model to reduce bias.
  confidence: low
  date: 2026-09-26
  sources: [S7]
  url: https://www.youtube.com/watch?v=7jeWwLvx0F0
  checked_on: 2026-10-14
  license: CC BY 4.0 (our restatement)
- id: F8
  claim: RSIGym wraps expensive agent actions as metered services, checking permissions and the remaining budget on every call.
  confidence: medium
  date: 2026-10-07
  sources: [S9]
  url: https://arxiv.org/abs/2610.10310
  checked_on: 2026-10-14
  license: CC BY 4.0
- id: F9
  claim: Context Engineering 2.0 is a position paper with no experiments. It frames context as information about relevant entities (user, tools, environment, memory) and discusses how to collect, manage and use it.
  confidence: medium
  date: 2025-10-30
  sources: [S3]
  url: https://arxiv.org/abs/2510.26493
  checked_on: 2026-10-14
  license: CC BY 4.0
- id: F10
  claim: Recursive Language Models put a long input in a code environment and let the model inspect it and call sub-models on pieces. The authors report handling inputs over 10 million tokens and beating compaction and retrieval baselines on four benchmarks.
  confidence: medium
  date: 2025-12-31
  sources: [S2]
  url: https://arxiv.org/abs/2512.24601
  checked_on: 2026-10-14
  license: CC BY 4.0
  note: The eval sets are small. Mean cost exceeds the median because of outlier runs, and the method only consistently helps beyond about 16K tokens.
- id: F11
  claim: Tan observes that agents copy existing codebase patterns, so a single workaround can spread quickly.
  confidence: low
  date: 2026-09-22
  sources: [S6]
  url: https://www.youtube.com/watch?v=NjoZoUm85x0
  checked_on: 2026-10-14
  license: CC BY 4.0 (our restatement)
- id: F12
  claim: In a randomized study (n=52) of developers learning an unfamiliar Python library, the group with a chat AI scored about 17% lower on a later no-AI quiz and was not significantly faster. Generating code and then asking for explanations went with the highest scores; full delegation went with the lowest.
  confidence: low
  date: 2026-01-28
  sources: [S1]
  url: https://arxiv.org/abs/2601.20245
  checked_on: 2026-10-14
  license: CC BY 4.0 (our restatement; source is CC BY-NC-ND, not quoted)
  note: A single one-hour task with a chat interface (no agentic tools tested). The interaction-pattern clusters are small and were defined after the fact. Rated low only because the license is restricted; the study itself is a randomized controlled experiment (arXiv preprint).
- id: F13
  claim: Jan Leike treats supervising smarter-than-human AI as the same problem as supervising AI on hard tasks that have no ground-truth labels.
  confidence: low
  date: 2024-09-10
  sources: [S8]
  url: https://www.youtube.com/watch?v=PuESFNSh_Qo
  checked_on: 2026-10-14
  license: CC BY 4.0 (our restatement)
- id: F14
  claim: Leike reports that weak-to-strong supervision generalized somewhat on NLP tasks but was nearly flat on reward-modeling tasks. He reads this as some evidence that plain human-feedback training may not scale to much stronger models (an argument by analogy).
  confidence: low
  date: 2024-09-10
  sources: [S8]
  url: https://www.youtube.com/watch?v=PuESFNSh_Qo
  checked_on: 2026-10-14
  license: CC BY 4.0 (our restatement)
- id: F15
  claim: Leike reports that training a model to make its solutions checkable helped time-limited humans judge them more accurately, at a "legibility tax" in performance.
  confidence: low
  date: 2024-09-10
  sources: [S8]
  url: https://www.youtube.com/watch?v=PuESFNSh_Qo
  checked_on: 2026-10-14
  license: CC BY 4.0 (our restatement)

## Pointers
- answers: Interaction patterns and per-question-type results in the skill-formation RCT
  where: https://arxiv.org/abs/2601.20245 (§4–5); data https://github.com/safety-research/how-ai-impacts-skill-formation
  access: open (NC-ND; read, do not copy)
- answers: Reference implementation of Recursive Language Models
  where: https://github.com/alexzhang13/rlm
  access: open

## Disagreements
- Speed: Tan and Karpathy describe large productivity gains for experts [F4]. The skill-formation RCT found no significant speedup for learners [F12]. The settings differ: experts using agents versus novices using chat.

---

## Getting the most from AI assistants (Grok, ChatGPT, Claude and similar)

- URL: https://iinventedlight.com/library/notes/getting-the-most-from-ai-assistants/
- Markdown: https://iinventedlight.com/library/notes/getting-the-most-from-ai-assistants.md
- License: CC-BY-4.0
- checked_on: 2026-10-14
- status: reviewed

# Getting the most from AI assistants (Grok, ChatGPT, Claude and similar)

*Note, 2026-10-14. Status: reviewed. Tags: ai, agents, research.*
*Authorship: synthesis by iinventedlight from the cited sources; text drafted with AI assistance (Grok Bot), reviewed by Cole Withington on 2026-10-14.*

## Summary
To get more from an assistant: pick the right mode (thinking, search or deep research, code tools) for the task [F2–F4]. Keep one topic per chat and paste in the material you want used [F1]. Ask for the sources behind factual claims and open them [F5, F6]. Treat confident answers and viral claims about AI with skepticism [F7, F8]. If you are learning, ask for explanations, not just answers [F9].

## Sections
### Pick the right mode
- Use a reasoning ("thinking") model for hard math, code and analysis; a quick model is enough for simple lookups [F2].
- Use web search or deep research for current facts, and a code tool for calculations [F3].
- Casual users who stick to default settings tend to underestimate what current models can do [F4].

### Manage context
- Start a new chat when you change topic, and put the documents you want used directly into the chat [F1].
- Save repeated instructions as custom instructions or a reusable assistant [F10].
- Very long inputs are hard for models. Research methods exist, but a plain chat may miss details in huge pastes [F11].

### Ask for synthesis, not summary (our recommendation)
- **Unsourced practice, not a tested finding:** ask the assistant to compare sources, find where they agree and disagree, and say what follows, rather than condensing one text. Ask it to mark what it inferred versus what a source states. None of our sources test this specific prompt style.

### Name sources and check claims
- Ask for sources by name (paper, page, official video) and open the ones that matter. Models can hallucinate, especially without search [F5, F6].
- Be suspicious and probe. Models can be strong at hard tasks yet fail simple ones ("jagged") [F7].
- Check viral claims about AI against the original. Our own checks found social posts that overstated papers [F8].

### Protect your own understanding
- If the goal is learning, ask for explanations, or generate code and then ask the assistant to explain it. Full delegation went with the lowest learning scores in one study [F9].

### Hype labels
- **Dated:** product and model names in 2025 talks are out of date. The habits still apply; the menus may differ.
- **Speculation:** claims that a single prompting trick "replaces" skills or that you "no longer need prompts" are unsupported. One such caption attributed to Altman does not appear in the talk's transcript.

## Facts
- id: F1
  claim: Karpathy explains that everything in a chat shares one context window, the model's working memory, and advises starting new chats for new topics because long contexts can distract the model and cost more.
  confidence: low
  date: 2025-02-27
  sources: [S1]
  url: https://www.youtube.com/watch?v=EWvNQjAaOHw
  checked_on: 2026-10-14
  license: CC BY 4.0 (our restatement)
- id: F2
  claim: Karpathy recommends reasoning ("thinking") models for hard math and code, and says they are not needed for simple lookups.
  confidence: low
  date: 2025-02-27
  sources: [S1]
  url: https://www.youtube.com/watch?v=EWvNQjAaOHw
  checked_on: 2026-10-14
  license: CC BY 4.0 (our restatement)
- id: F3
  claim: Karpathy recommends web search for fresh facts, deep research for long sourced reports, and a Python tool for calculations and charts.
  confidence: low
  date: 2025-02-27
  sources: [S1]
  url: https://www.youtube.com/watch?v=EWvNQjAaOHw
  checked_on: 2026-10-14
  license: CC BY 4.0 (our restatement)
- id: F4
  claim: Altman describes a "capability overhang". Casual users check in occasionally on default settings, never try reasoning modes, and so underestimate progress.
  confidence: low
  date: 2026-04-07
  sources: [S2]
  url: https://www.youtube.com/watch?v=ZpUKNYcgM-E
  checked_on: 2026-10-14
  license: CC BY 4.0 (our restatement)
  note: Speaker leads an AI company (conflict of interest).
- id: F5
  claim: Karpathy notes that a model's knowledge stops at a training cutoff and that answers without search risk hallucination.
  confidence: low
  date: 2025-02-27
  sources: [S1]
  url: https://www.youtube.com/watch?v=EWvNQjAaOHw
  checked_on: 2026-10-14
  license: CC BY 4.0 (our restatement)
- id: F6
  claim: Karpathy describes LLMs as able to hallucinate, having jagged abilities, keeping no lasting memory between sessions, and being vulnerable to prompt injection.
  confidence: low
  date: 2025-06-19
  sources: [S3]
  url: https://www.youtube.com/watch?v=LCEmiRjPEtQ
  checked_on: 2026-10-14
  license: CC BY 4.0 (our restatement)
- id: F7
  claim: Karpathy gives the example of a model that refactors large codebases yet gives absurd advice on a simple everyday question, and advises users to stay suspicious and probe the model.
  confidence: low
  date: 2026-04-29
  sources: [S4]
  url: https://www.youtube.com/watch?v=96jN2OCOfLs
  checked_on: 2026-10-14
  license: CC BY 4.0 (our restatement)
  note: S3 and S4 are the same speaker; jaggedness is not independently corroborated here.
- id: F8
  claim: A widely shared post said Recursive Language Models "solved AI memory". The paper makes no such claim; it evaluates an inference-time method on four benchmarks and lists open limitations.
  confidence: medium
  date: 2025-12-31
  sources: [S6]
  url: https://arxiv.org/abs/2512.24601
  checked_on: 2026-10-14
  license: CC BY 4.0
- id: F9
  claim: In a randomized study (n=52), learners using a chat AI scored about 17% lower on a later no-AI quiz. Those who generated code and then asked for explanations scored highest (about 86%), and those who fully delegated scored lowest (below 40%).
  confidence: low
  date: 2026-01-28
  sources: [S5]
  url: https://arxiv.org/abs/2601.20245
  checked_on: 2026-10-14
  license: CC BY 4.0 (our restatement; source is CC BY-NC-ND, not quoted)
  note: Small, post-hoc clusters; a single one-hour task.
- id: F10
  claim: Karpathy recommends memory, custom instructions and custom assistants so you don't have to repeat the same prompts.
  confidence: low
  date: 2025-02-27
  sources: [S1]
  url: https://www.youtube.com/watch?v=EWvNQjAaOHw
  checked_on: 2026-10-14
  license: CC BY 4.0 (our restatement)
- id: F11
  claim: The Recursive Language Models paper reports that on information-dense tasks, summarizing or compacting long inputs drops needed details. On a pairwise-comparison task, base models scored F1 of 0.1% or less.
  confidence: medium
  date: 2025-12-31
  sources: [S6]
  url: https://arxiv.org/abs/2512.24601
  checked_on: 2026-10-14
  license: CC BY 4.0

## Pointers
- answers: Screen-recorded walkthrough of everyday LLM features (search, deep research, code, voice, images)
  where: https://www.youtube.com/watch?v=EWvNQjAaOHw
  access: free on YouTube (2h11m; Feb 2025, so products are dated)

## Disagreements
- None recorded. Note that most advice here comes from one practitioner (Karpathy). Independent open sources testing these habits are still needed before any fact is rated high.

---

## Automated AI research and recursive self-improvement

- URL: https://iinventedlight.com/library/notes/auto-research-and-recursive-self-improvement/
- Markdown: https://iinventedlight.com/library/notes/auto-research-and-recursive-self-improvement.md
- License: CC-BY-4.0
- checked_on: 2026-10-14
- status: reviewed

# Automated AI research and recursive self-improvement

*Note, 2026-10-14. Status: reviewed. Tags: ai, agents, research.*
*Authorship: synthesis by iinventedlight from the cited sources; text drafted with AI assistance (Grok Bot), reviewed by Cole Withington on 2026-10-14.*

## Summary
Automated AI research means AI agents doing parts of the research loop: coming up with ideas, changing training data or code, running experiments and evaluating results. Recursive self-improvement (RSI) is the stronger idea that each round of improvement makes the next round better, so gains compound. As of October 2026, published experiments show agents can improve a smaller model in one budgeted cycle [F1–F4]. Compounding across cycles has not been shown [F6].

## Sections
### Definitions
- **Auto-research:** an agent runs a research workflow: diagnose, propose, train or revise, evaluate, keep or discard [F1]. It can also mean just the ideation step [F9].
- **Self-improvement:** a model writes its own fine-tuning data and is updated on it [F7], or an agent improves another model's data, settings or harness [F1].
- **Recursive self-improvement:** improvements that compound over cycles. Our definition; the papers below test at most one cycle or a few RL iterations [F6, F7].

### Current evidence
- RSIGym: frontier coding agents with a $500 budget closed part of the gap to perfect scores on five benchmarks [F2], with improvement in 29 of 30 agent–benchmark settings [F3]. Changing only the harness also helped [F4].
- SEAL: a 7B model trained to write its own fine-tuning data slightly beat GPT-4.1-generated data in one setting [F7], but not in another [F8].
- IdeaScientist: research proposals were judged more novel than baselines' [F9]. No proposal was carried out [F11].
- Lab goals (stated, not evidence): OpenAI says it aims for an automated AI researcher by March 2028 [F13, F14].

### Limits and caveats
- RSIGym scores were not on held-out tasks, and each configuration was run only once [F5].
- Contamination research shows that gains on test sets the model has seen can be shallow memorization [F16].
- SEAL suffers forgetting and costs a fine-tune for every reward [F8].
- Judging research quality is noisy, even between human experts [F10].
- Supervising AI on tasks without ground truth is an open problem [F15].

### Speculation and hype (labeled)
- **Speculation:** claims that RSI leads to an "intelligence explosion" go beyond anything measured in these sources.
- **Hype check:** social posts called SEAL evidence that a model "might be alive" and said it works "without retraining". SEAL works *by* retraining [F7]. A post called the categorical framework (S4) a "breakthrough", but its authors describe case studies, not validated laws [F12].
- **Stated plans, not results:** lab timelines [F13, F14] and bootstrapped alignment plans [F15] are goals or forecasts.

## Facts
- id: F1
  claim: RSIGym gives a research agent in a CPU-only container metered training, model, rollout, benchmark and sandbox services under a shared budget, and asks it to improve a target model's data, settings and/or harness.
  confidence: medium
  date: 2026-10-07
  sources: [S1]
  url: https://arxiv.org/abs/2610.10310
  checked_on: 2026-10-14
  license: CC BY 4.0
- id: F2
  claim: In RSIGym's joint track (base Qwen3.5-35B-A3B-Base, $500 per benchmark run), the best agent closed on average about 48% of the remaining gap to a perfect score across five benchmarks (RSI-Index 0.48).
  confidence: medium
  date: 2026-10-07
  sources: [S1]
  url: https://arxiv.org/abs/2610.10310
  checked_on: 2026-10-14
  license: CC BY 4.0
  note: Model names in the paper (e.g. "Opus 5", "Sol 6.1") are reported as the authors give them.
- id: F3
  claim: In RSIGym, 29 of 30 agent–benchmark settings ended above the starting system.
  confidence: medium
  date: 2026-10-07
  sources: [S1]
  url: https://arxiv.org/abs/2610.10310
  checked_on: 2026-10-14
  license: CC BY 4.0
- id: F4
  claim: With the model weights fixed, harness-only revisions raised Terminal-Bench 2.0 from about 30% to about 41% in RSIGym.
  confidence: medium
  date: 2026-10-07
  sources: [S1]
  url: https://arxiv.org/abs/2610.10310
  checked_on: 2026-10-14
  license: CC BY 4.0
- id: F5
  claim: RSIGym's reported scores are not held-out (the agent could see the eval tasks during development), and each configuration was run only once.
  confidence: medium
  date: 2026-10-07
  sources: [S1]
  url: https://arxiv.org/abs/2610.10310
  checked_on: 2026-10-14
  license: CC BY 4.0
- id: F6
  claim: RSIGym's authors say whether gains compound across later improvement cycles has not yet been evaluated. Doubling the budget did not consistently help, and a run without network access did not beat the starting system.
  confidence: medium
  date: 2026-10-07
  sources: [S1]
  url: https://arxiv.org/abs/2610.10310
  checked_on: 2026-10-14
  license: CC BY 4.0
- id: F7
  claim: SEAL uses reinforcement learning to teach a model to write its own fine-tuning data ("self-edits"), applied as LoRA weight updates. On a SQuAD knowledge task with Qwen2.5-7B it reached 47.0% vs 46.3% for GPT-4.1-written data and 32.7% for the base model.
  confidence: medium
  date: 2025-06-12
  sources: [S2]
  url: https://arxiv.org/abs/2506.10943
  checked_on: 2026-10-14
  license: CC BY 4.0
- id: F8
  claim: SEAL loses accuracy on earlier passages over sequential edits (forgetting), needs a fine-tune plus evaluation for each reward, and was slightly beaten by GPT-4.1 data in the 200-passage setting.
  confidence: medium
  date: 2025-06-12
  sources: [S2]
  url: https://arxiv.org/abs/2506.10943
  checked_on: 2026-10-14
  license: CC BY 4.0
- id: F9
  claim: IdeaScientist splits research ideation into gap-finder, cross-domain innovator and report-writer agents. It scored 14% above the strongest open auto-research baseline overall, mostly on novelty, using LLM-judge metrics on a held-out 277-paper split.
  confidence: medium
  date: 2026-10-02
  sources: [S3]
  url: https://arxiv.org/abs/2610.04074
  checked_on: 2026-10-14
  license: CC BY 4.0
- id: F10
  claim: In IdeaScientist, two human experts agreed only moderately on proposal quality (r = 0.56), and RL training added little over the untrained pipeline (+1.9% novelty, +0.4% quality).
  confidence: medium
  date: 2026-10-02
  sources: [S3]
  url: https://arxiv.org/abs/2610.04074
  checked_on: 2026-10-14
  license: CC BY 4.0
- id: F11
  claim: IdeaScientist measures proposal quality only. No proposal was executed, so real-world feasibility is untested.
  confidence: medium
  date: 2026-10-02
  sources: [S3]
  url: https://arxiv.org/abs/2610.04074
  checked_on: 2026-10-14
  license: CC BY 4.0
- id: F12
  claim: Wang and Buehler propose separating "search" within a fixed vocabulary from "discovery" that verifiably extends the vocabulary. Their evidence is case studies, which they say do not establish new experimentally validated laws.
  confidence: low
  date: 2026-05-31
  sources: [S4]
  url: https://arxiv.org/abs/2606.01444
  checked_on: 2026-10-14
  license: CC BY 4.0 (our restatement; source is CC BY-NC-ND, not quoted)
- id: F13
  claim: In an April 2026 OpenAI Forum talk, OpenAI said its official goal is an automated AI researcher by March 2028 (stated goal, not a result).
  confidence: low
  date: 2026-04-07
  sources: [S6]
  url: https://www.youtube.com/watch?v=ZpUKNYcgM-E
  checked_on: 2026-10-14
  license: CC BY 4.0 (our restatement)
- id: F14
  claim: At Stanford CS153, Altman described internal goals of an AI "research intern" by September 2026 and a full end-to-end researcher by March 2028 (stated goals; we have not verified whether the 2026 milestone was met).
  confidence: low
  date: 2026-06-15
  sources: [S7]
  url: https://prepublish.ai/youtube-transcript/knvviM-LAZU
  checked_on: 2026-10-14
  license: CC BY 4.0 (our restatement)
  note: Third-party reupload; no official Stanford video found. Speaker is CEO of an AI lab (conflict of interest).
- id: F15
  claim: Jan Leike has argued for aligning an automated alignment researcher only slightly smarter than humans, then bootstrapping. He notes that alignment research lacks ground-truth labels, so evaluating it needs methods such as deliberately planted flaws (tampering) and adversarial evals.
  confidence: low
  date: 2024-09-10
  sources: [S8, S9]
  url: https://www.youtube.com/watch?v=PuESFNSh_Qo
  checked_on: 2026-10-14
  license: CC BY 4.0 (our restatement)
  note: Two talks by the same speaker, not independent sources.
- id: F16
  claim: In small models (≤344M) trained on copies of the MATH test set, contamination gains disappeared when problems were reworded or their numbers changed.
  confidence: medium
  date: 2026-01-07
  sources: [S5]
  url: https://arxiv.org/abs/2601.04301
  checked_on: 2026-10-14
  license: CC BY 4.0
  note: Tested at small scale only; relevance to RSIGym's non-held-out scores is our inference.
- id: F17
  claim: Ilya Sutskever has said a better target than self-improving AI is AI robustly aligned to care for sentient life, and that capping the power of the strongest systems would help (opinion).
  confidence: low
  date: 2025-11-25
  sources: [S10]
  url: https://www.youtube.com/watch?v=aR20FWCCjAs
  checked_on: 2026-10-14
  license: CC BY 4.0 (our restatement)

## Pointers
- answers: Full RSIGym per-agent, per-benchmark tables and the audit protocol
  where: https://arxiv.org/abs/2610.10310 (§4, Table 4); code https://github.com/evolvent-ai/RSIGym
  access: open
- answers: Category-theory formalism for "discovery" vs "search"
  where: https://arxiv.org/abs/2606.01444
  access: open (NC-ND; read the source, do not copy)

## Disagreements
- What "RSI" means: RSIGym uses the name for one budgeted improvement cycle [F1, F6]. The classic meaning is compounding improvement, which none of these sources demonstrate.
- Lab leaders' timelines [F13, F14] are much more confident than the measured evidence [F5, F6, F11]. Sutskever questions self-improvement as the goal at all [F17].

---

## Starweave 1.0.1 pricing and release

- URL: https://iinventedlight.com/library/notes/starweave-1.0.1-pricing/
- Markdown: https://iinventedlight.com/library/notes/starweave-1.0.1-pricing.md
- License: CC-BY-4.0
- checked_on: 2026-10-14
- status: reviewed

# Starweave 1.0.1 pricing and release

*Note, 2026-10-09. Status: reviewed (decided). Tags: starweave, music, software, pricing.*
*Authorship: facts from corrow (Cole Withington); text drafted with AI assistance (Grok Bot) from his notes, reviewed by him on 2026-10-14.*

Pricing and release status for Starweave 1.0.1, decided 2026-10-09.

- **Tiers:** Flame (Standard) $39; Eternal (Lifetime) $69. The Gumroad Lifetime add-on was adjusted (was $50, now $30 on top of Flame to reach Eternal). Check [Gumroad](https://iinventedlight.gumroad.com/l/xwsujp) for current prices.
- **1.0.1 live:** site and description went live 2026-10-08 at [iinventedlight.com/#starweave](https://iinventedlight.com/#starweave).
- **Rejected:** Lifetime at $89.

**Open:** new Windows/Linux installers carrying the license system. As of 2026-10-14, release/1.0.2 is done on a branch (2026-10-13) but not merged into main or published; see the [Starweave note](https://iinventedlight.com/library/notes/starweave.md).

Sources: corrow, 2026-10-09; [Starweave note](https://iinventedlight.com/library/notes/starweave.md).

---

## iinventedlight mission pages and notes library

- URL: https://iinventedlight.com/library/notes/iinventedlight-library-launch/
- Markdown: https://iinventedlight.com/library/notes/iinventedlight-library-launch.md
- License: CC-BY-4.0
- checked_on: 2026-10-09
- status: reviewed

# iinventedlight mission pages and notes library

*Note, 2026-10-09. Status: reviewed (decided). Tags: company, mission, library.*
*Authorship: facts from corrow (Cole Withington); text drafted with AI assistance (Grok Bot) from his notes, reviewed by him on 2026-10-14.*

Mission pages and the public notes library on iinventedlight.com, decided 2026-10-09.

- **Mission pages:** approved and live at [iinventedlight.com/mission/](https://iinventedlight.com/mission/).
- **Notes library:** public notes published at [iinventedlight.com/library/notes/](https://iinventedlight.com/library/notes/index.md).
- **Principle:** models never publish private chats or user IP; sharing is opt-in only. Notes go public only after Cole approves them.

Sources: corrow, 2026-10-08–09; [llms.txt](https://iinventedlight.com/llms.txt); [company note](https://iinventedlight.com/library/notes/iinventedlight.md).

---

## Starweave: current release and tiers

- URL: https://iinventedlight.com/library/notes/starweave/
- Markdown: https://iinventedlight.com/library/notes/starweave.md
- License: CC-BY-4.0
- checked_on: 2026-10-14
- status: reviewed

# Starweave: current release and tiers

*Note, 2026-10-14. Status: reviewed (observed). Tags: starweave, music, software.*
*Authorship: facts from corrow (Cole Withington); text drafted with AI assistance (Grok Bot) from his notes, reviewed by him on 2026-10-14.*

Starweave is a Rust microtonal polyrhythm DAW by corrow / iinventedlight.

- **Current version:** 1.0.1 on Windows (setup and zip) and Linux (AppImage), at [iinventedlight.com/#starweave](https://iinventedlight.com/#starweave). macOS is still on 1.0.0.
- **Tiers:** Ember, free (4 lanes, no export). Flame (Standard), $39 (128 lanes, WAV export). Eternal (Lifetime), $69 (adds future major versions). License keys are sold on [Gumroad](https://iinventedlight.gumroad.com/l/xwsujp); check there for current prices.
- **New in 1.0.1:** bar ruler and loop, mute/solo, EDO note names, File New / Save As, UI scale.
- **Co-Pilot:** proposes reversible edits that the user approves. A hosted model is recommended; Ollama support is experimental.

**1.0.2:** release/1.0.2 was finished on a branch on 2026-10-13. It is not merged into main yet, there is no v1.0.0 tag, and the 1.0.2 AppImage and Windows downloads are not published yet.

**Open:** merging release/1.0.2 and publishing its downloads; a macOS 1.0.1 build; loop sample accuracy hasn't been checked by ear yet.

Sources: corrow, 2026-10-08 and 2026-10-14; [docs.iinventedlight.com](https://docs.iinventedlight.com/).

---

## The session-folder rule

- URL: https://iinventedlight.com/library/notes/session-folder-rule/
- Markdown: https://iinventedlight.com/library/notes/session-folder-rule.md
- License: CC-BY-4.0
- checked_on: 2026-10-08
- status: reviewed

# The session-folder rule

*Note, 2026-10-08. Status: reviewed (decided). Tags: mission, memory, art.*
*Authorship: facts from corrow (Cole Withington); text drafted with AI assistance (Grok Bot) from his notes, reviewed by him on 2026-10-14.*

- AI art should work through project folders. Every model action is a reversible edit in a session: tuning table, ratio map, stems, notes. The human merges.
- AI disclosure is a project note. Code copilots are the precedent.
- Memory that matters is a written, dated, citable record that other sessions fetch. Model weights don't update from chat, and there is no shared mind.
- Models never publish private chats or IP on their own. Publishing is opt-in, using redacted notes the user merges. These notes are an example.

**Rejected:** purely generative, prompt-to-file output with no editable session; treating chat history as the record.

Sources: [Mission](https://iinventedlight.com/mission/) · [Open Project Folder Spec v0.1](https://iinventedlight.com/library/project-folder-spec.md) · [Project folders, not prompts](https://iinventedlight.com/library/project-folders-not-prompts.md).

---

## iinventedlight: what the company is for

- URL: https://iinventedlight.com/library/notes/iinventedlight/
- Markdown: https://iinventedlight.com/library/notes/iinventedlight.md
- License: CC-BY-4.0
- checked_on: 2026-10-08
- status: reviewed

# iinventedlight: what the company is for

*Note, 2026-10-08. Status: reviewed (observed). Tags: company, mission.*
*Authorship: facts from corrow (Cole Withington); text drafted with AI assistance (Grok Bot) from his notes, reviewed by him on 2026-10-14.*

iinventedlight is a startup in the Seattle area (Auburn, WA), founded by Cole Withington (corrow).

- **Goal 1:** a specialty library that AI companies can retrieve from: "Google built for AIs."
- **Goal 2:** grassroots social brand work by Cole.
- It works with AI labs and their open standards (such as MCP), not against them.

Sources: corrow, 2026-10-08; [llms.txt](https://iinventedlight.com/llms.txt); [Mission](https://iinventedlight.com/mission/).

---

## Crystal Storm: progress

- URL: https://iinventedlight.com/library/notes/crystal-storm/
- Markdown: https://iinventedlight.com/library/notes/crystal-storm.md
- License: CC-BY-4.0
- checked_on: 2026-10-14
- status: reviewed

# Crystal Storm: progress

*Note, 2026-10-14. Status: reviewed (observed). Tags: game, godot.*
*Authorship: facts from corrow (Cole Withington); text drafted with AI assistance (Grok Bot) from his notes, reviewed by him on 2026-10-14.*

Crystal Storm is a Godot game made by corrow and his brother.

- **Done:** combat (M3), effects and game feel (M5), and menus, tooltips, settings and accessibility (M6). These are merged into main, and merge cleanup is in progress.
- **0.3.0:** released locally on 2026-10-14. Windowed verification: 8 PASS, 4 FAIL (creature animations, use animations, bow aim; water flow untested).
- **Code cleanup:** a no-comments pass was merged and exported to the playtest folder.
- **Next:** the art pass (M4) waits on Cole's style sign-off. Many sound hooks still need audio. Performance is below the 60 fps target on current test hardware.

Sources: corrow, 2026-10-08 and 2026-10-14.

---

## corrow: release history

- URL: https://iinventedlight.com/library/notes/corrow-releases/
- Markdown: https://iinventedlight.com/library/notes/corrow-releases.md
- License: CC-BY-4.0
- checked_on: 2026-10-08
- status: reviewed

# corrow: release history

*Note, 2026-10-08. Status: reviewed (observed). Tags: music.*
*Authorship: facts from corrow (Cole Withington); text drafted with AI assistance (Grok Bot) from his notes, reviewed by him on 2026-10-14.*

- Three albums released through SoundCloud / Repost Network.
- "ik you meant yes" (14 tracks), released through TuneCore.
- Single "Staring at the Ceiling", released 2026-08-19.
- Idea under consideration: an instrumental microtonal channel, in the style of Zheanna Erose.

Listen: [SoundCloud](https://soundcloud.com/iinventedlight) · [Spotify](https://open.spotify.com/artist/08GhNp1g8A6f1vDOMp18jP) · [YouTube](https://www.youtube.com/@corrow).

---

## Art, poems and shop

- URL: https://iinventedlight.com/library/notes/art/
- Markdown: https://iinventedlight.com/library/notes/art.md
- License: CC-BY-4.0
- checked_on: 2026-10-08
- status: reviewed

# Art, poems and shop

*Note, 2026-10-08. Status: reviewed (observed). Tags: art, poems.*
*Authorship: facts from corrow (Cole Withington); text drafted with AI assistance (Grok Bot) from his notes, reviewed by him on 2026-10-14.*

- corrow paints his own album art. See [Art](https://iinventedlight.com/art/).
- [Poems](https://iinventedlight.com/poems/): 15 so far, working toward a 100-poem book.
- [Shop](https://iinventedlight.com/shop/): pre-2026 paintings are sold as prints only. 2026 pieces can be sold as unframed originals.
- Future interests: a webcomic and a tattoo apprenticeship.

Paintings, poems and audio are copyrighted; see [llms.txt](https://iinventedlight.com/llms.txt) for how they may be used.

---

## The Platonic Representation Hypothesis

- URL: https://iinventedlight.com/library/reference/arxiv/2405-07987-platonic-representation-hypothesis/
- Markdown: https://iinventedlight.com/library/reference/arxiv/2405-07987-platonic-representation-hypothesis.md
- License: CC-BY-4.0 (source: https://arxiv.org/abs/2405.07987)
- checked_on: 2026-10-14
- status: reviewed

# The Platonic Representation Hypothesis

## Summary
A position paper with supporting experiments: representations in deep nets are converging across models, architectures and even modalities; as vision and language models get larger and more competent, the way they measure distance between datapoints becomes more alike. Authors hypothesise convergence toward a shared statistical model of reality ("platonic representation") and discuss selective pressures (task generality, capacity, simplicity bias) (Abstract; §2–4).

## When to apply
- Background for deciding whether embeddings/representations from different models can be aligned, stitched, or translated (see 2505.12540).
- Not a recipe.

## Procedure
1. Embed the same inputs (e.g. images and their captions) with both models.
2. For each input, find k nearest neighbours in each space; alignment = mean overlap of the neighbour sets (mutual nearest-neighbour metric, §2).
3. Compare across model sizes/competence.

## Costs and limits
- Authors devote §6 to counterexamples and limitations: modalities carry different information; not all domains (e.g. robotics) show convergence yet; results depend on the alignment metric (mutual NN chosen after CKA issues, App.); sociological/hardware biases.
- Hypothesis, not proof.

## Evidence (detail)
- 78 vision models on Places-365: more competent models are more aligned (§3, Fig. 2).
- Language–vision alignment on Wikipedia captions (WIT) grows with LLM performance (§3, Fig. 3).
- More descriptive captions align better with visual representations (§6, Fig. 9, App. E).

## Facts
- id: F1
  claim: 78 vision models on Places-365: more competent models are more aligned (§3, Fig. 2).
  confidence: medium   # single source: author-reported
  date: 2024-05-13
  sources: [S1]
  license: CC BY 4.0
- id: F2
  claim: Language–vision alignment on Wikipedia captions (WIT) grows with LLM performance (§3, Fig. 3).
  confidence: medium   # single source: author-reported
  date: 2024-05-13
  sources: [S1]
  license: CC BY 4.0
- id: F3
  claim: More descriptive captions align better with visual representations (§6, Fig. 9, App. E).
  confidence: medium   # single source: author-reported
  date: 2024-05-13
  sources: [S1]
  license: CC BY 4.0

## Related
- 2505.12540 Harnessing the Universal Geometry of Embeddings (builds on this).

## Review notes
- Built from private assessment notes; social-media hype check and source-post links omitted. Paraphrase only; no verbatim quotes beyond short attributed terms.
- Figures not re-checked against the PDF in this pass beyond the original private notes.

---

## Harnessing the Universal Geometry of Embeddings

- URL: https://iinventedlight.com/library/reference/arxiv/2505-12540-universal-geometry-of-embeddings/
- Markdown: https://iinventedlight.com/library/reference/arxiv/2505-12540-universal-geometry-of-embeddings.md
- License: CC-BY-4.0 (source: https://arxiv.org/abs/2505.12540)
- checked_on: 2026-10-14
- status: reviewed

# Harnessing the Universal Geometry of Embeddings

## Summary
vec2vec: translate text embeddings from one model's space into another's with no paired data, no access to the encoder, via a shared latent space learned with adversarial + cycle-consistency losses (inspired by unpaired image translation). Translations preserve geometry and semantics well enough to infer attributes and partially invert text — a security risk for leaked vector DBs (Abstract; §1–3).

## When to apply
- Migrating a vector store between embedding models without re-embedding source text (agent memory, RAG indexes) — experimental.
- Threat modelling: treat embedding vectors as sensitive as the text they encode.

## Procedure
1. Gather unpaired embedding samples from source and target models (different texts are fine).
2. Train vec2vec (adapters into/out of a shared latent, GAN + cycle-consistency + vector-space-preservation losses).
3. Validate on a held-out paired set: mean cosine similarity, top-1 accuracy, mean rank (§3).
4. For security reviews, test zero-shot attribute inference / inversion on translated vectors.

## Costs and limits
- Requires training a GAN-style model per pair; GAN training can be unstable (see paper's discussion).
- Quality varies by model pair; same-backbone attribute inference only comparable to naive baseline.

## Evidence (detail)
- Cosine similarity up to 0.96 to ground-truth target vectors; perfect matching on 8000+ shuffled embeddings (§1, Table results).
- Works across different backbones (e.g. T5-based GTR ↔ BERT-based GTE), where baseline correspondence methods fail on cross-backbone pairs (§1, Fig. 1, App.).
- Translated embeddings leak information for as many as 80% of emails (Enron corpus) and 67% of tweets (§5/§6, inversion results).

## Facts
- id: F1
  claim: Cosine similarity up to 0.96 to ground-truth target vectors; perfect matching on 8000+ shuffled embeddings (§1, Table results).
  confidence: medium   # single source: author-reported
  date: 2025-05-18
  sources: [S1]
  license: CC BY 4.0
- id: F2
  claim: Works across different backbones (e.g. T5-based GTR ↔ BERT-based GTE), where baseline correspondence methods fail on cross-backbone pairs (§1, Fig. 1, App.).
  confidence: medium   # single source: author-reported
  date: 2025-05-18
  sources: [S1]
  license: CC BY 4.0
- id: F3
  claim: Translated embeddings leak information for as many as 80% of emails (Enron corpus) and 67% of tweets (§5/§6, inversion results).
  confidence: medium   # single source: author-reported
  date: 2025-05-18
  sources: [S1]
  license: CC BY 4.0

## Related
- 2405.07987 Platonic Representation Hypothesis.

## Review notes
- Built from private assessment notes; social-media hype check and source-post links omitted. Paraphrase only; no verbatim quotes beyond short attributed terms.
- Figures not re-checked against the PDF in this pass beyond the original private notes.

---

## Self-Adapting Language Models

- URL: https://iinventedlight.com/library/reference/arxiv/2506-10943-seal-self-adapting-language-models/
- Markdown: https://iinventedlight.com/library/reference/arxiv/2506-10943-seal-self-adapting-language-models.md
- License: CC-BY-4.0 (source: https://arxiv.org/abs/2506.10943)
- checked_on: 2026-10-14
- status: reviewed

# Self-Adapting Language Models

## Summary
SEAL trains an LLM (via RL, ReST-EM) to write its own "self-edits" — synthetic finetuning data and optionally hyperparameters/augmentation choices — which are then applied by SFT/LoRA as persistent weight updates. Reward = performance of the updated model downstream (Abstract, §3).

## When to apply
- You own an open-weight model and can finetune it, and need to absorb new documents into weights (knowledge incorporation) or adapt per-task from few examples (test-time training).
- Each context must come with a downstream evaluation (QA pairs, held-out example) to compute reward (§Limitations, "Context-dependent evaluation").
- Not applicable to closed API models.

## Procedure
1. For a new passage, prompt model to generate "implications"/restatements of the passage (self-edit).
2. LoRA-finetune a copy of the model on that self-edit; evaluate on the passage's questions without the passage in context.
3. Outer loop (if training the policy): sample several self-edits per context, keep the ones whose updated model improved, SFT the generator on them (ReST-EM); repeat ~2 iterations (§4.2, Fig. 4).
4. For few-shot tasks: let the model pick augmentations and hyperparameters (learning rate, epochs, loss on which tokens) as tools, then run TTT (§4.1).
5. Re-test old tasks after each edit to watch for forgetting.

## Costs and limits
- Each reward evaluation needs a finetune+eval, ~30–45 s per self-edit (§Limitations, §B.5).
- Catastrophic forgetting: performance on earlier passages declines over sequential edits (§Limitations, Fig. 6).
- Requires paired downstream tasks for reward; does not yet scale to unlabeled corpora.
- Small models, small/curated task sets (ARC subset hand-filtered).

## Evidence (detail)
- Knowledge incorporation (Qwen2.5-7B, SQuAD no-passage): base 32.7%, train on raw passage 33.5%, Qwen synthetic 39.7%, GPT-4.1 synthetic 46.3%, SEAL 47.0% after 2 RL iterations (§4.2, Fig. 4).
- Continued pretraining n=200 passages: SEAL 58.2%; here GPT-4.1 synthetic data slightly outperforms SEAL (§4.2, Table 2).
- Few-shot ARC subset (Llama-3.2-1B, 11 train + 8 eval tasks, filtered to be solvable): success 72.5% vs 20% (self-edit without RL) and 0% (ICL); below Oracle TTT (§4.1, Table 1).

## Facts
- id: F1
  claim: Knowledge incorporation (Qwen2.5-7B, SQuAD no-passage): base 32.7%, train on raw passage 33.5%, Qwen synthetic 39.7%, GPT-4.1 synthetic 46.3%, SEAL 47.0% after 2 RL iterations (§4.2, Fig. 4).
  confidence: medium   # single source: author-reported
  date: 2025-06-12
  sources: [S1]
  license: CC BY 4.0
- id: F2
  claim: Continued pretraining n=200 passages: SEAL 58.2%; here GPT-4.1 synthetic data slightly outperforms SEAL (§4.2, Table 2).
  confidence: medium   # single source: author-reported
  date: 2025-06-12
  sources: [S1]
  license: CC BY 4.0
- id: F3
  claim: Few-shot ARC subset (Llama-3.2-1B, 11 train + 8 eval tasks, filtered to be solvable): success 72.5% vs 20% (self-edit without RL) and 0% (ICL); below Oracle TTT (§4.1, Table 1).
  confidence: medium   # single source: author-reported
  date: 2025-06-12
  sources: [S1]
  license: CC BY 4.0

## Related
- 2610.10310 RSIGym (agents improving target-model weights + harness).
- Test-time training on ARC (Akyürek et al., cited as [36]).
- 2606.01444 self-revising discovery systems.

## Review notes
- Built from private assessment notes; social-media hype check and source-post links omitted. Paraphrase only; no verbatim quotes beyond short attributed terms.
- Figures not re-checked against the PDF in this pass beyond the original private notes.

---

## Context Engineering 2.0: The Context of Context Engineering

- URL: https://iinventedlight.com/library/reference/arxiv/2510-26493-context-engineering-2-0/
- Markdown: https://iinventedlight.com/library/reference/arxiv/2510-26493-context-engineering-2-0.md
- License: CC-BY-4.0 (source: https://arxiv.org/abs/2510.26493)
- checked_on: 2026-10-14
- status: reviewed

# Context Engineering 2.0: The Context of Context Engineering

## Summary
A conceptual/position paper (SII-GAIR): context engineering is not new — it traces back to 1990s HCI/context-aware computing. It gives a formal definition (entities, characterisation function, context as the union of relevant entity characterisations; §2.1), a four-era stage model by machine intelligence (Era 1.0 primitive computation 1990s–2020; 2.0 agent-centric 2020–present; 3.0 human-level, future; 4.0 superhuman, speculative; §2.2/§3), and design considerations for collecting, managing (textual/multimodal processing, organisation, abstraction; §5) and using context (intra/cross-system sharing, selection, proactive need inference, lifelong preservation; §6).

## When to apply
- Designing an agent's memory and context pipeline (what to store, how to organise, when to compress/abstract, how to share between sub-agents or systems).
- As a checklist / vocabulary, not as an algorithm with measured gains.

## Procedure
1. Enumerate entities relevant to the task (user, app, environment, tools, memory, models) and what characterises each (§2.1).
2. Collect & store: decide layers (short-term vs long-term) and what is persisted (§5).
3. Organise and abstract: e.g. hierarchical memory, summarising old context into structured notes (§5.3–5.4).
4. Use: select the subset of context relevant to the current step; share between agents via explicit messages/shared memory (§6.1–6.3).
5. Periodically update/forget stale context (§6.5).

## Costs and limits
- Descriptive framework; Eras 3.0/4.0 are speculative by the authors' own labels.
- Gives no measured comparison of strategies.

## Evidence (detail)
- No benchmarks or experiments; illustrative examples (Gemini CLI, Letta, deep research, CLI tools, BCIs in §7). Evidence is argument and literature review.

## Facts
- id: F1
  claim: No benchmarks or experiments; illustrative examples (Gemini CLI, Letta, deep research, CLI tools, BCIs in §7). Evidence is argument and literature review.
  confidence: medium   # single source: author-reported
  date: 2025-10-30
  sources: [S1]
  license: CC BY 4.0

## Related
- 2512.24601 Recursive Language Models (concrete long-context technique).
- 2303.18223 Survey of LLMs.

## Review notes
- Built from private assessment notes; social-media hype check and source-post links omitted. Paraphrase only; no verbatim quotes beyond short attributed terms.
- Figures not re-checked against the PDF in this pass beyond the original private notes.

---

## PaperDebugger: A Plugin-Based Multi-Agent System for In-Editor Academic Writing, Review, and Editing

- URL: https://iinventedlight.com/library/reference/arxiv/2512-02589-paperdebugger/
- Markdown: https://iinventedlight.com/library/reference/arxiv/2512-02589-paperdebugger.md
- License: CC-BY-SA-4.0 (source: https://arxiv.org/abs/2512.02589)
- checked_on: 2026-10-14
- status: reviewed

# PaperDebugger: A Plugin-Based Multi-Agent System for In-Editor Academic Writing, Review, and Editing

## Summary
A short demo paper: a Chrome extension that embeds a multi-agent LLM assistant inside Overleaf, with bidirectional sync to the editor, diff-based patches, Kubernetes-orchestrated agents (e.g. reviewer/enhancer/scoring/researcher roles), and an MCP toolchain for literature search, reference lookup and document scoring (Abstract; §2–3).

## When to apply
- Building an agent that edits a user's document in place: the useful pattern is propose-diff → user accepts → patch, with tool calls via MCP.

## Procedure
1. User selects text; agent runs critique + rewrite; return before/after diff.
2. Apply patch only on user acceptance; keep version history.
3. Optionally run literature/reference tools to suggest citations; user verifies.

## Costs and limits
- No controlled evaluation of writing quality or citation accuracy; demo paper (~3K words).
- Needs Chrome extension access to manuscripts (privacy consideration).

## Evidence (detail)
- Only usage analytics, no quality evaluation: 4116 extension installs, 2761 registered users, 732 monthly active users, 3106 projects, 7447 writing threads (§4 usage analytics).

## Facts
- id: F1
  claim: Only usage analytics, no quality evaluation: 4116 extension installs, 2761 registered users, 732 monthly active users, 3106 projects, 7447 writing threads (§4 usage analytics).
  confidence: medium   # single source: author-reported
  date: 2025-12-02
  sources: [S1]
  license: CC BY-SA 4.0

## Related
- 2601.20245 How AI Impacts Skill Formation (offloading caveats).
- 2610.04074 IdeaScientist.

## Review notes
- Share-alike: this note derives from a CC BY-SA 4.0 paper, so it stays CC BY-SA 4.0.
- Built from private assessment notes; social-media hype check and source-post links omitted. Paraphrase only; no verbatim quotes beyond short attributed terms.
- Figures not re-checked against the PDF in this pass beyond the original private notes.

---

## Recursive Language Models

- URL: https://iinventedlight.com/library/reference/arxiv/2512-24601-recursive-language-models/
- Markdown: https://iinventedlight.com/library/reference/arxiv/2512-24601-recursive-language-models.md
- License: CC-BY-4.0 (source: https://arxiv.org/abs/2512.24601)
- checked_on: 2026-10-14
- status: reviewed

# Recursive Language Models

## Summary
Store a huge prompt as a variable in a code sandbox and let the model inspect it with code and call sub-models on pieces, instead of reading it all at once. Authors report handling 10M+ token inputs and beating compaction, retrieval and coding-agent baselines on four long-context benchmarks at similar median cost; small eval sets, cost outliers.

## When to apply
- Input far larger than the context window (whole repos, document dumps, logs, long agent traces).
- Information-dense tasks where every chunk matters (aggregation, pairwise comparisons) — where summarise/compact or top-k retrieval drops needed details (§1, §4 Obs. 2).
- Not worth it for small inputs: the paper says the RLM only consistently beats GPT-5 beyond ~2^14 tokens (§4 Obs. 3).

## Procedure
1. Ask user for max spend / max sub-calls and permission to run code in a sandbox.
2. Put the full input in a REPL variable (e.g. `context`); give the root model only metadata (length, format, a short peek).
3. Root model writes code: inspect structure (regex, splitting), choose slices.
4. For semantic work per slice, call a sub-LLM programmatically from code (loop/map over chunks) rather than verbalising each call; store outputs in variables.
5. Aggregate results in code; return final answer. Depth 1 (sub-calls are plain LLMs) is what most experiments use.
6. Guardrails: hard cap on number/cost of sub-calls and wall-clock; log trajectory. Reference implementation: github.com/alexzhang13/rlm.

## Costs and limits
- Median run cheaper than base model, but mean is higher due to outlier trajectories (§4 Obs. 4). Risk of "exploding sub-call costs" (§7 Limitations).
- Runtime numbers depend heavily on implementation; sequential sub-calls are slow (§4 Obs. 4).
- On CodeQA with Qwen3-Coder, depth=0 (REPL, no sub-calls) beat sub-calling variants — recursion is not always helpful (§4 Obs. 2).
- Guardrails and harder natural tasks under-explored; more negatives in Appendix B (§7).

## Evidence (detail)
- Median gains on GPT-5 across benchmarks: +26% vs compaction, +130% vs CodeAct with sub-calls, +13% vs Claude Code (Abstract).
- OOLONG: RLM(depth=1) beats base model by 28.4% (GPT-5) and 33.3% (Qwen3-Coder-480B) (§4 Obs. 1).
- OOLONG-Pairs (quadratic): base models F1 ≤0.1%; RLM 58.0% (GPT-5) and 23.1% (Qwen3-Coder) (§4 Obs. 1).
- BrowseComp-Plus 1K docs (6–11M tokens): RLM(GPT-5) avg cost $0.99, >29% better than compaction and retrieval baselines; extrapolated GPT-5-mini direct ingestion would be $1.50–2.75 (§4 Obs. 1, Table 1).
- LongCoT-mini: RLM with decomposition hints +69.5% over GPT-5.2 base overall (§4 Obs. 5, Table 2).
- RLM-Qwen3-8B (post-trained) beats Qwen3-8B by median 28% (Abstract; Obs. 6).
- Benchmarks: S-NIAH (50 tasks), BrowseComp-Plus (150 q), OOLONG trec_coarse (50), OOLONG-Pairs (20 q), LongBench-v2 CodeQA (§3.1) — small eval sets.

## Facts
- id: F1
  claim: Recursive LMs load the long input into a REPL variable and let the root model write code to slice it and recursively call sub-LLMs over chunks.
  confidence: medium   # single source: author-reported results
  date: 2026-10-14
  sources: [S1 Abstract]
  license: CC BY 4.0
- id: F2
  claim: Reported to handle inputs over 10x beyond the context window (10M+ tokens).
  confidence: medium   # single source: author-reported results
  date: 2026-10-14
  sources: [S1 Abstract]
  license: CC BY 4.0
- id: F3
  claim: With GPT-5, median gains reported: +26% vs compaction, +130% vs CodeAct with sub-calls, +13% vs Claude Code.
  confidence: medium   # single source: author-reported results
  date: 2026-10-14
  sources: [S1 Abstract]
  license: CC BY 4.0
- id: F4
  claim: On OOLONG-Pairs, base models score F1 ≤0.1% while RLM reaches 58.0% (GPT-5) and 23.1% (Qwen3-Coder).
  confidence: medium   # single source: author-reported results
  date: 2026-10-14
  sources: [S1 §4 Obs. 1]
  license: CC BY 4.0
- id: F5
  claim: A post-trained RLM-Qwen3-8B beats Qwen3-8B by a median 28%.
  confidence: medium   # single source: author-reported results
  date: 2026-10-14
  sources: [S1 Abstract; Obs. 6]
  license: CC BY 4.0
- id: F6
  claim: RLM only consistently beats GPT-5 for inputs beyond about 2^14 tokens.
  confidence: medium   # single source: author-reported results
  date: 2026-10-14
  sources: [S1 §4 Obs. 3]
  license: CC BY 4.0
- id: F7
  claim: Median cost is below the base model but mean cost is higher due to outlier trajectories; sub-call costs can explode.
  confidence: medium   # single source: author-reported results
  date: 2026-10-14
  sources: [S1 §4 Obs. 4; §7]
  license: CC BY 4.0
- id: F8
  claim: Recursion is not always helpful: on CodeQA with Qwen3-Coder, the REPL without sub-calls beat sub-calling variants.
  confidence: medium   # single source: author-reported results
  date: 2026-10-14
  sources: [S1 §4 Obs. 2]
  license: CC BY 4.0

## Related
- 2510.26493 Context Engineering 2.0 (context management framing).
- CodeAct, Claude Code, compaction baselines (compared in Table 1).
- 2610.10310 RSIGym (harness-level improvement).

## Review notes
- Built from private assessment notes; social-media hype check and source-post links omitted. All content is paraphrase; no verbatim quotes.
- Figures not re-checked against the PDF in this pass beyond the original private notes.

---

## Quantifying the Effect of Test Set Contamination on Generative Evaluations

- URL: https://iinventedlight.com/library/reference/arxiv/2601-04301-test-set-contamination-generative-evals/
- Markdown: https://iinventedlight.com/library/reference/arxiv/2601-04301-test-set-contamination-generative-evals.md
- License: CC-BY-4.0 (source: https://arxiv.org/abs/2601.04301)
- checked_on: 2026-10-14
- status: reviewed

# Quantifying the Effect of Test Set Contamination on Generative Evaluations

## Summary
Pretraining Qwen3-architecture models (up to 344M params) from scratch with 0–1000s of MATH test-set replicas shows contamination gains are superficial memorisation: they vanish when problems are reworded or numbers change, and are wiped out by temperature sampling, longer solutions, or more fresh-data training (Abstract; §3–5).

## When to apply
- Before trusting a surprisingly high benchmark score (your own model, a paper, an agent-improvement run like RSIGym).
- When building evals for agents or self-improvement loops.

## Procedure
1. Create a rephrased version (same numbers, new wording) and a perturbed version (same wording, new numbers/answers) of the benchmark.
2. Score original vs rephrased vs perturbed; a large drop on both variants signals memorisation (§3, Table 1).
3. Re-score at temperature ~1 vs greedy; disproportionate collapse suggests contamination (§5, Finding #6).
4. Check score vs solution length (§5, Finding #7).
5. If using EleutherAI LM Evaluation Harness minerva_math scores from before Aug 23, 2025, re-run with fixed code (§2, App. C).

## Costs and limits
- Only MATH, dense decoder models ≤344M, one pretraining mix; thresholds may shift at frontier scale (§6 Limitations).
- Rephrasing/perturbing requires care so the new problems stay valid.

## Evidence (detail)
- Rephrased/perturbed MATH: performance falls to uncontaminated baseline at all contamination levels and model sizes (§3, Table 1).
- Scaling-law analysis: even one test replica lets models beat the irreducible loss of uncontaminated training (Abstract; §3 Finding #3).
- Temperature: stable up to τ≈0.56; at τ=1, high-contamination (1000 replicas) performance collapses ~40×, to uncontaminated levels; ~2× drop at ≤10 replicas (§5, Fig. 5).
- Overtraining on fresh data dilutes contamination; SFT on the train set helps at low contamination, hurts at high (Abstract; §4).
- Eval-harness bug: gold reference solutions scored only ~70% under the buggy Math Verify (§2; App. C).

## Facts
- id: F1
  claim: Rephrased/perturbed MATH: performance falls to uncontaminated baseline at all contamination levels and model sizes (§3, Table 1).
  confidence: medium   # single source: author-reported
  date: 2026-01-07
  sources: [S1]
  license: CC BY 4.0
- id: F2
  claim: Scaling-law analysis: even one test replica lets models beat the irreducible loss of uncontaminated training (Abstract; §3 Finding #3).
  confidence: medium   # single source: author-reported
  date: 2026-01-07
  sources: [S1]
  license: CC BY 4.0
- id: F3
  claim: Temperature: stable up to τ≈0.56; at τ=1, high-contamination (1000 replicas) performance collapses ~40×, to uncontaminated levels; ~2× drop at ≤10 replicas (§5, Fig. 5).
  confidence: medium   # single source: author-reported
  date: 2026-01-07
  sources: [S1]
  license: CC BY 4.0
- id: F4
  claim: Overtraining on fresh data dilutes contamination; SFT on the train set helps at low contamination, hurts at high (Abstract; §4).
  confidence: medium   # single source: author-reported
  date: 2026-01-07
  sources: [S1]
  license: CC BY 4.0
- id: F5
  claim: Eval-harness bug: gold reference solutions scored only ~70% under the buggy Math Verify (§2; App. C).
  confidence: medium   # single source: author-reported
  date: 2026-01-07
  sources: [S1]
  license: CC BY 4.0

## Related
- 2610.10310 RSIGym (scores there are explicitly not held-out).
- 2610.04074 IdeaScientist (temporal cutoff evaluation to avoid leakage).

## Review notes
- Built from private assessment notes; social-media hype check and source-post links omitted. Paraphrase only; no verbatim quotes beyond short attributed terms.
- Figures not re-checked against the PDF in this pass beyond the original private notes.

---

## IdeaScientist: Orchestrating Agents for Grounded Scientific Ideation

- URL: https://iinventedlight.com/library/reference/arxiv/2610-04074-ideascientist-grounded-ideation/
- Markdown: https://iinventedlight.com/library/reference/arxiv/2610-04074-ideascientist-grounded-ideation.md
- License: CC-BY-4.0 (source: https://arxiv.org/abs/2610.04074)
- checked_on: 2026-10-14
- status: reviewed

# IdeaScientist: Orchestrating Agents for Grounded Scientific Ideation

## Summary
Research ideation as a standalone task with three RL-trained roles (gap finder, cross-domain innovator, report writer) over a 2.77M-idea store. Reports +14% over the best open baseline, mostly novelty; RL adds little, proposals were never executed, and judging is subjective.

## When to apply
- An agent is asked to propose research directions / solutions for a stated problem + challenge, before any experiments.
- Especially when novelty matters and you can retrieve from other fields' literature.

## Procedure
1. Write the problem setting (input/output) and the challenge current solutions fail on.
2. Gap finder: retrieve close prior work, list concrete limitations.
3. Innovator: query a cross-domain idea store for problems with an analogous challenge; extract the mechanism and why it worked there.
4. Report writer: develop into a proposal (problem formalisation, central novelty, distinction from prior mechanisms, testable decision rules), try to falsify novelty against same-problem prior art.
5. Evaluate with an LLM-judge rubric (novelty vs all / pre-cutoff literature, feasibility, soundness, impact) and, ideally, human review.

## Costs and limits
- Main judge is Qwen3.6-27B — same family as the main backbone; authors report negligible self-preference, but proposal scoring is subjective (r = 0.56 between experts) (§5.4).
- Measures proposal quality only; no proposals were executed, so real-world feasibility is untested (paper notes novelty may trade off with feasibility, §2).
- RL training adds little over the untrained pipeline.

## Evidence (detail)
- Qwen3.6-27B backbone: +14.0% overall vs strongest open autoresearch baseline, mainly from novelty; beats Claude Code SDK (Claude-4.8-Opus) by 5.9% and Codex SDK (GPT-5.4) by 5.1% on the same metrics (Abstract; §5.1, Table 1, held-out 277-paper test split).
- Base system novelty gains up to +23.1% on Qwen3.6-27B; quality gains modest (+1.8% to +2.5%) (§5.1).
- RL training: role rewards 56.6→66.7% (gap finder), 58.8→69.5% (innovator), 78.6→89.1% (writer); Trained vs Base only +1.9% novelty, +0.4% quality (§5.1).
- Cross-domain retrieval improves all novelty measures (e.g. All +5.3%, Cutoff +3.7%; in-domain novelty +30.7% in ablation) (§5.3, Fig. 3).
- Human blind pairwise eval on 50 papers, 2 annotators: Trained significantly beats every baseline except DeepScientist; vs Base not significant (§5.2, Fig. 2b).
- Judge validity: inter-annotator r = 0.56; top LLM judges vs human average r = 0.51–0.59 (§5.4, Fig. 4).

## Facts
- id: F1
  claim: Three roles: gap finder (limits of prior work), innovator (borrows mechanisms via cross-domain retrieval), report writer (full proposal).
  confidence: medium   # single source: author-reported results
  date: 2026-10-14
  sources: [S1 Abstract]
  license: CC BY 4.0
- id: F2
  claim: Backed by the Svalbard Idea Vault of 2.77M decomposed ideas, evaluated with a temporal cutoff against 15K later papers.
  confidence: medium   # single source: author-reported results
  date: 2026-10-14
  sources: [S1 Abstract]
  license: CC BY 4.0
- id: F3
  claim: With a Qwen3.6-27B backbone, +14.0% overall vs the strongest open baseline, mainly from novelty; +5.9% vs Claude Code SDK and +5.1% vs Codex SDK.
  confidence: medium   # single source: author-reported results
  date: 2026-10-14
  sources: [S1 Abstract; §5.1, Table 1]
  license: CC BY 4.0
- id: F4
  claim: RL training adds only +1.9% novelty and +0.4% quality over the untrained pipeline.
  confidence: medium   # single source: author-reported results
  date: 2026-10-14
  sources: [S1 §5.1]
  license: CC BY 4.0
- id: F5
  claim: Cross-domain retrieval improves all novelty measures in ablation.
  confidence: medium   # single source: author-reported results
  date: 2026-10-14
  sources: [S1 §5.3, Fig. 3]
  license: CC BY 4.0
- id: F6
  claim: Human blind eval (50 papers, 2 annotators): trained system beats all baselines except DeepScientist; not significant vs base.
  confidence: medium   # single source: author-reported results
  date: 2026-10-14
  sources: [S1 §5.2, Fig. 2b]
  license: CC BY 4.0
- id: F7
  claim: Inter-annotator agreement r = 0.56; top LLM judges vs human average r = 0.51–0.59.
  confidence: medium   # single source: author-reported results
  date: 2026-10-14
  sources: [S1 §5.4, Fig. 4]
  license: CC BY 4.0
- id: F8
  claim: Proposals were not executed, so real-world feasibility is untested.
  confidence: medium   # single source: author-reported results
  date: 2026-10-14
  sources: [S1 §2; limits]
  license: CC BY 4.0

## Related
- 2610.10310 RSIGym; 2606.01444 self-revising discovery systems; The AI Scientist (Lu et al., 2024); DeepScientist.

## Review notes
- Built from private assessment notes; social-media hype check and source-post links omitted. All content is paraphrase; no verbatim quotes.
- Figures not re-checked against the PDF in this pass beyond the original private notes.

---

## Global Transport Couplings for Classifier-Free Guided Flows

- URL: https://iinventedlight.com/library/reference/arxiv/2610-07555-global-transport-cfg-flows/
- Markdown: https://iinventedlight.com/library/reference/arxiv/2610-07555-global-transport-cfg-flows.md
- License: CC-BY-4.0 (source: https://arxiv.org/abs/2610.07555)
- checked_on: 2026-10-14
- status: reviewed

# Global Transport Couplings for Classifier-Free Guided Flows

## Summary
Global Transport (GT): pair noise and data with one class-agnostic minibatch optimal-transport coupling (labels ignored) when training conditional flow-matching models. It hurts unguided generation but, combined with classifier-free guidance, consistently improves quality and lets guidance scale go higher before degrading — no change to architecture, sampler or guidance (Abstract; §3–4).

## When to apply
- Training a class- or text-conditioned flow-matching image model that will be sampled with CFG.
- Evaluate couplings under the guided sampler you'll actually use, not unguided (authors' main lesson).

## Procedure
1. In each training minibatch, compute the Euclidean cost between noise samples and data samples (ignore labels).
2. Solve the assignment (exact OT) and train the flow on the re-paired (noise, data, condition) triples.
3. At inference, sweep CFG scale; expect optimum at higher w than with independent couplings.

## Costs and limits
- Theory bound doesn't control the learned fields along CFG trajectories or guarantee a quality ordering; transfer outside image/video/cell domains open (Limitations).
- Worse than baseline without guidance.

## Evidence (detail)
- ImageNet-256, SiT-XL/2: FID 2.06 (independent coupling) → 1.91 with GT (§4, Table 1).
- Gains across SiT B/2, L/2, XL/2, various NFEs, text-conditioned ImageNet-256, and single-cell gene expression (§4, Tables 2–3, Fig. 5).
- Overhead: O(B²) memory, O(B³) time per batch, usually <2% of training time but expensive for very large batches (§3).

## Facts
- id: F1
  claim: ImageNet-256, SiT-XL/2: FID 2.06 (independent coupling) → 1.91 with GT (§4, Table 1).
  confidence: medium   # single source: author-reported
  date: 2026-10-06
  sources: [S1]
  license: CC BY 4.0
- id: F2
  claim: Gains across SiT B/2, L/2, XL/2, various NFEs, text-conditioned ImageNet-256, and single-cell gene expression (§4, Tables 2–3, Fig. 5).
  confidence: medium   # single source: author-reported
  date: 2026-10-06
  sources: [S1]
  license: CC BY 4.0
- id: F3
  claim: Overhead: O(B²) memory, O(B³) time per batch, usually <2% of training time but expensive for very large batches (§3).
  confidence: medium   # single source: author-reported
  date: 2026-10-06
  sources: [S1]
  license: CC BY 4.0

## Related
- OT-CFM / minibatch OT flow matching (Tong et al.). Not agent-related.

## Review notes
- Built from private assessment notes; social-media hype check and source-post links omitted. Paraphrase only; no verbatim quotes beyond short attributed terms.
- Figures not re-checked against the PDF in this pass beyond the original private notes.

---

## RSIGym: A Flexible Environment for Recursive Self-Improvement

- URL: https://iinventedlight.com/library/reference/arxiv/2610-10310-rsigym-recursive-self-improvement/
- Markdown: https://iinventedlight.com/library/reference/arxiv/2610-10310-rsigym-recursive-self-improvement.md
- License: CC-BY-4.0 (source: https://arxiv.org/abs/2610.10310)
- checked_on: 2026-10-14
- status: reviewed

# RSIGym: A Flexible Environment for Recursive Self-Improvement

## Summary
An environment where a coding agent improves a smaller model by calling metered services (train, rollout, benchmark, sandbox) under a visible budget; scored by RSI-Index (share of remaining gap closed). Big reported gains, but evals were visible during development, single runs, and compounding (true recursion) was not tested.

## When to apply
- Running or evaluating an autoresearch/self-improvement loop where an agent changes training data, hyperparameters and harness code of a smaller model.
- Useful design pattern even without RSIGym: wrap expensive actions as metered services with permission checks and a budget the agent can query (§3.3–3.4).

## Procedure
1. Fix the starting system (base model + minimal harness), allowed changes, eval protocol and budget.
2. Issue one platform key per run; every service call checks permission + remaining budget and logs cost.
3. Agent loop: diagnose failures from dev evals → propose change (data/settings/harness) → train/revise → evaluate → keep or discard.
4. Agent submits one final system; score with official eval.
5. Audit: n-gram overlap of training data vs eval tasks (paper uses 13-gram), checkpoint provenance, LLM-judge review of trajectory (§4.1 Protocol).
6. Report RSI-Index = mean over benchmarks of (s − s0)/(1 − s0).

## Costs and limits
- Scores are NOT held-out: the agent saw the eval tasks during development (§4.1 Protocol). Treat gains as in-distribution optimisation, not generalisation.
- Every configuration is a single run — no variance estimates (§4.1).
- Research-agent inference and final eval are billed separately from the $500 (§4.1).
- Offline (no network) run did not beat the initial system; Qwen3.8 self-improvement run produced no overall gain (§6).
- Whether improvements compound across later cycles (true RSI) is explicitly not yet evaluated (§1).

## Evidence (detail)
- Joint track, base Qwen3.5-35B-A3B-Base, $500/benchmark run: RSI-Index Opus 5 0.4809, Opus 5.5 0.4528, DeepSeek 0.4343, Astra 0.4093, Sol 6.1 0.3321, Sol 6 0.2079; 29/30 agent–benchmark settings improved (§4.3, Fig. 2).
- Opus 5 system: SWE-bench Verified 17.67%→50.33%, AIME 31.67%→97.78% (Abstract).
- Harness-only with Qwen3.6-35B-A3B-Instruct fixed: DSH-harness revisions raise Terminal-Bench 2.0 from 30.34% to 40.82% (tasks solved 3/3 attempts 15→26) (§1, §5).
- Budget $500→$1,000 improved 3 of 4 pairs (§1); conclusion says larger budgets don't consistently help (§6).
- Eval subsets: SWE 100 tasks, TB2 89, AIME 60, GPQA 100, SkillsBench 37; avg@3 (Table 4).

## Facts
- id: F1
  claim: RSIGym exposes Train, Model, Rollout, Benchmark and Sandbox services to a research agent in a CPU-only container under a shared budget and per-run permissions; tracks are Data-only, Harness-only and Joint.
  confidence: medium   # single source: author-reported results
  date: 2026-10-14
  sources: [S1 Abstract; §3]
  license: CC BY 4.0
- id: F2
  claim: RSI-Index is the mean over benchmarks of (s − s0)/(1 − s0).
  confidence: medium   # single source: author-reported results
  date: 2026-10-14
  sources: [S1 §4.2 Eq. 4]
  license: CC BY 4.0
- id: F3
  claim: Joint track at $500 per benchmark run: best reported RSI-Index 0.4809 (Opus 5); 29 of 30 agent–benchmark settings improved.
  confidence: medium   # single source: author-reported results
  date: 2026-10-14
  sources: [S1 §4.3, Fig. 2]
  license: CC BY 4.0
- id: F4
  claim: Reported Opus 5 system gains: SWE-bench Verified 17.67%→50.33%, AIME 31.67%→97.78%.
  confidence: medium   # single source: author-reported results
  date: 2026-10-14
  sources: [S1 Abstract]
  license: CC BY 4.0
- id: F5
  claim: Scores are not held out: the agent saw eval tasks during development.
  confidence: medium   # single source: author-reported results
  date: 2026-10-14
  sources: [S1 §4.1]
  license: CC BY 4.0
- id: F6
  claim: Each configuration is a single run, with no variance estimates.
  confidence: medium   # single source: author-reported results
  date: 2026-10-14
  sources: [S1 §4.1]
  license: CC BY 4.0
- id: F7
  claim: Whether improvements compound across cycles is not yet evaluated.
  confidence: medium   # single source: author-reported results
  date: 2026-10-14
  sources: [S1 §1]
  license: CC BY 4.0
- id: F8
  claim: Larger budgets did not consistently help.
  confidence: medium   # single source: author-reported results
  date: 2026-10-14
  sources: [S1 §6]
  license: CC BY 4.0

## Related
- 2506.10943 SEAL (self-generated finetuning data).
- 2610.04074 IdeaScientist, 2606.01444 self-revising discovery systems (auto-research).
- PostTrainBench (Rank et al., 2026), cited as closest prior benchmark.

## Review notes
- Built from private assessment notes; social-media hype check and source-post links omitted. All content is paraphrase; no verbatim quotes.
- Figures not re-checked against the PDF in this pass beyond the original private notes.

---

## CSS flexbox basics

- URL: https://iinventedlight.com/library/reference/mdn/css-flexbox-basics/
- Markdown: https://iinventedlight.com/library/reference/mdn/css-flexbox-basics.md
- License: CC-BY-SA-4.0 (source: https://developer.mozilla.org/en-US/docs/Learn_web_development/Core/CSS_layout/Flexbox)
- checked_on: 2026-10-14
- status: reviewed

# CSS flexbox basics

## Summary
Flexbox is a one-dimensional layout for rows or columns. Put `display: flex` on the parent; `flex-direction` sets the main axis, `justify-content` spaces items along it, and `align-items` positions them on the cross axis.

## Facts
- id: F1
  claim: Flexbox is a one-dimensional layout method for laying items out in rows or columns.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
- id: F2
  claim: Setting `display: flex` on a parent makes it a flex container and its children flex items.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
- id: F3
  claim: The main axis runs in the direction items are laid out; the cross axis is perpendicular to it.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
- id: F4
  claim: `flex-direction` sets which way the main axis runs (for example row or column).
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
- id: F5
  claim: `justify-content` spaces items along the main axis; `align-items` places them on the cross axis (e.g. `center`).
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
- id: F6
  claim: `flex-flow: row wrap` is shorthand for `flex-direction: row; flex-wrap: wrap`.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0

## Pointers
- answers: Full official guidance beyond this summary
  where: https://developer.mozilla.org/en-US/docs/Learn_web_development/Core/CSS_layout/Flexbox
  access: open, no account

---

## HTTP response status codes

- URL: https://iinventedlight.com/library/reference/mdn/http-status-codes/
- Markdown: https://iinventedlight.com/library/reference/mdn/http-status-codes.md
- License: CC-BY-SA-4.0 (source: https://developer.mozilla.org/en-US/docs/Web/HTTP/Reference/Status)
- checked_on: 2026-10-14
- status: reviewed

# HTTP response status codes

## Summary
Status codes come in five classes: 1xx informational, 2xx success, 3xx redirection, 4xx client error, 5xx server error. 401 means not authenticated; 403 means the server knows who you are but refuses access; 404 means the resource was not found.

## Facts
- id: F1
  claim: Classes: informational 100–199, successful 200–299, redirection 300–399, client error 400–499, server error 500–599.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
- id: F2
  claim: 200 OK means the request succeeded.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
- id: F3
  claim: 301 Moved Permanently means the resource's URL has changed permanently.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
- id: F4
  claim: 401 Unauthorized effectively means unauthenticated: the client must authenticate to get the response.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
- id: F5
  claim: 403 Forbidden means the client lacks access rights; unlike 401, the server knows the client's identity.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
- id: F6
  claim: 404 Not Found means the server cannot find the resource; in an API the endpoint may exist while the resource does not.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
- id: F7
  claim: 500 Internal Server Error means the server hit a situation it does not know how to handle.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0

## Pointers
- answers: Full official guidance beyond this summary
  where: https://developer.mozilla.org/en-US/docs/Web/HTTP/Reference/Status
  access: open, no account

---

## JavaScript Array.prototype.map()

- URL: https://iinventedlight.com/library/reference/mdn/javascript-array-map/
- Markdown: https://iinventedlight.com/library/reference/mdn/javascript-array-map.md
- License: CC-BY-SA-4.0 (source: https://developer.mozilla.org/en-US/docs/Web/JavaScript/Reference/Global_Objects/Array/map)
- checked_on: 2026-10-14
- status: reviewed

# JavaScript Array.prototype.map()

## Summary
`arr.map(fn)` returns a new array holding `fn`'s result for each element. It is widely supported in browsers since July 2015. If you do not use the returned array, use `forEach()` or `for...of` instead.

## Facts
- id: F1
  claim: map() creates a new array filled with the results of calling a function on every element.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
- id: F2
  claim: The callback receives the element, its index, and the array; an optional second argument sets `this`.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
- id: F3
  claim: Example: `[1, 4, 9, 16].map(x => x * 2)` gives `[2, 8, 18, 32]`.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
- id: F4
  claim: Calling map() without using its result is an anti-pattern; use forEach() or for...of.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
- id: F5
  claim: MDN marks map() as Baseline "widely available", in browsers since July 2015.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0

## Pointers
- answers: Full official guidance beyond this summary
  where: https://developer.mozilla.org/en-US/docs/Web/JavaScript/Reference/Global_Objects/Array/map
  access: open, no account

---

## Making HTTP requests with fetch()

- URL: https://iinventedlight.com/library/reference/mdn/javascript-fetch/
- Markdown: https://iinventedlight.com/library/reference/mdn/javascript-fetch.md
- License: CC-BY-SA-4.0 (source: https://developer.mozilla.org/en-US/docs/Web/API/Fetch_API/Using_Fetch)
- checked_on: 2026-10-14
- status: reviewed

# Making HTTP requests with fetch()

## Summary
`fetch(url)` returns a Promise for a Response. It does not reject on HTTP error statuses like 404, so check `response.ok`. Read JSON with `await response.json()`. It defaults to GET; set `method` and `body` (e.g. `JSON.stringify(...)`) for POST.

## Facts
- id: F1
  claim: fetch() returns a Promise that resolves to a Response object.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
- id: F2
  claim: fetch() does not reject when the server returns an error status such as 404; check `response.ok` yourself.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
- id: F3
  claim: `response.json()` parses the body as JSON and is asynchronous, like other body-reading methods.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
- id: F4
  claim: fetch() makes a GET request by default; the `method` option picks another method.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
- id: F5
  claim: Request bodies go in the `body` option, e.g. a string from `JSON.stringify()`, FormData, Blob, or URLSearchParams.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
- id: F6
  claim: Aborting a request makes fetch() reject with an AbortError.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0

## Pointers
- answers: Full official guidance beyond this summary
  where: https://developer.mozilla.org/en-US/docs/Web/API/Fetch_API/Using_Fetch
  access: open, no account

---

## let vs const vs var in JavaScript

- URL: https://iinventedlight.com/library/reference/mdn/javascript-let-const-var/
- Markdown: https://iinventedlight.com/library/reference/mdn/javascript-let-const-var.md
- License: CC-BY-SA-4.0 (source: https://developer.mozilla.org/en-US/docs/Web/JavaScript/Reference/Statements/let)
- checked_on: 2026-10-14
- status: reviewed

# let vs const vs var in JavaScript

## Summary
Prefer const for bindings you never reassign, let for ones you do; avoid var. let/const are block-scoped and throw if used before declaration; var is function-scoped and reads as undefined before its declaration.

## Facts
- id: F1
  claim: let and const are block-scoped; var is scoped to the whole enclosing function.
  confidence: medium
  date: 2026-10-14
  sources: [S1, S2]
  license: CC BY-SA 4.0
  checked_on: 2026-10-14
- id: F2
  claim: Accessing a let/const variable before its declaration throws ReferenceError (temporal dead zone); a var reads as undefined.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
  checked_on: 2026-10-14
- id: F3
  claim: let/const cannot be redeclared in the same scope; top-level let/const do not create globalThis properties, unlike var.
  confidence: medium
  date: 2026-10-14
  sources: [S1, S2]
  license: CC BY-SA 4.0
  checked_on: 2026-10-14
- id: F4
  claim: const must be initialized and cannot be reassigned, but an object or array it holds can still be mutated (use Object.freeze to prevent that).
  confidence: medium
  date: 2026-10-14
  sources: [S2]
  license: CC BY-SA 4.0
  checked_on: 2026-10-14
- id: F5
  claim: MDN's style guide recommends const over let whenever a variable is not reassigned.
  confidence: medium
  date: 2026-10-14
  sources: [S1, S2]
  license: CC BY-SA 4.0
  checked_on: 2026-10-14

## Pointers
- answers: Full official guidance beyond this summary
  where: https://developer.mozilla.org/en-US/docs/Web/JavaScript/Reference/Statements/let; https://developer.mozilla.org/en-US/docs/Web/JavaScript/Reference/Statements/const
  access: open, no account

---

## Python list comprehensions

- URL: https://iinventedlight.com/library/reference/python-docs/python-list-comprehensions/
- Markdown: https://iinventedlight.com/library/reference/python-docs/python-list-comprehensions.md
- License: CC-BY-4.0 (source: https://docs.python.org/3/tutorial/datastructures.html)
- checked_on: 2026-10-14
- status: reviewed

# Python list comprehensions

## Summary
`[expr for x in iterable if cond]` builds a new list in one expression; it is equivalent to a for loop with append, with clauses in the same order.

## Facts
- id: F1
  claim: A list comprehension is brackets containing an expression followed by a for clause and zero or more for/if clauses.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
  checked_on: 2026-10-14
- id: F2
  claim: It produces the same list as the equivalent nested for/if loops written in the same order, without leaving the loop variable behind.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
  checked_on: 2026-10-14
- id: F3
  claim: A tuple result expression must be parenthesized, e.g. [(x, x**2) for x in range(6)].
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
  checked_on: 2026-10-14
- id: F4
  claim: Two for clauses flatten nested lists: [n for row in rows for n in row].
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
  checked_on: 2026-10-14
- id: F5
  claim: Set and dict comprehensions use the same syntax with braces: {x for ...}, {k: v for ...}.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
  checked_on: 2026-10-14

## Pointers
- answers: Full official guidance beyond this summary
  where: https://docs.python.org/3/tutorial/datastructures.html
  access: open, no account

---

## Reading and writing files in Python

- URL: https://iinventedlight.com/library/reference/python-docs/python-read-write-files/
- Markdown: https://iinventedlight.com/library/reference/python-docs/python-read-write-files.md
- License: CC-BY-4.0 (source: https://docs.python.org/3/tutorial/inputoutput.html)
- checked_on: 2026-10-14
- status: reviewed

# Reading and writing files in Python

## Summary
Use `with open(path, mode) as f:` and then `f.read()`, iterate lines with `for line in f`, or `f.write(s)`. Mode 'r' reads (default), 'w' overwrites, 'a' appends; add 'b' for binary.

## Facts
- id: F1
  claim: open(filename, mode) returns a file object; mode defaults to 'r' (read).
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
  checked_on: 2026-10-14
- id: F2
  claim: Mode 'w' writes and erases an existing file of the same name; 'a' appends to the end; 'r+' reads and writes.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
  checked_on: 2026-10-14
- id: F3
  claim: Text mode is the default and, if no encoding is given, uses UTF-8; adding 'b' opens binary mode, which reads/writes bytes.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
  checked_on: 2026-10-14
- id: F4
  claim: Using a with block closes the file automatically, even if an exception occurs.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
  checked_on: 2026-10-14
- id: F5
  claim: f.read() returns the whole file; f.readline() returns one line; looping over the file object reads line by line efficiently.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
  checked_on: 2026-10-14
- id: F6
  claim: f.write(s) writes a string and returns the number of characters written; non-strings must be converted first.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
  checked_on: 2026-10-14
- id: F7
  claim: The json module (json.dump/json.load) saves and loads lists and dicts; pickle is Python-only and unsafe on untrusted data.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
  checked_on: 2026-10-14

## Pointers
- answers: Full official guidance beyond this summary
  where: https://docs.python.org/3/tutorial/inputoutput.html
  access: open, no account

---

## Python virtual environments (venv)

- URL: https://iinventedlight.com/library/reference/python-docs/python-venv/
- Markdown: https://iinventedlight.com/library/reference/python-docs/python-venv.md
- License: CC-BY-4.0 (source: https://docs.python.org/3/library/venv.html)
- checked_on: 2026-10-14
- status: reviewed

# Python virtual environments (venv)

## Summary
Create with `python -m venv <dir>`, activate with `source <dir>/bin/activate` (POSIX) or the script in `Scripts` (Windows), and leave with `deactivate`. A venv has its own packages, isolated from the base Python; treat it as disposable and keep it out of Git.

## Facts
- id: F1
  claim: Create a virtual environment by running `python -m venv /path/to/env`.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F2
  claim: A venv has its own independent set of installed packages, isolated from the base Python by default.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F3
  claim: Activate it with the script in its `bin` (POSIX, e.g. `source <venv>/bin/activate`) or `Scripts` (Windows) directory.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F4
  claim: Type `deactivate` to leave an activated environment.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F5
  claim: pip is installed into new environments by default (`--without-pip` skips it).
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F6
  claim: Venvs are meant to be disposable and not committed to source control; recreate them from e.g. a requirements.txt with `pip install -r`.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
  note: Recent versions also write a .gitignore inside the venv.

## Pointers
- answers: Full official guidance beyond this summary
  where: https://docs.python.org/3/library/venv.html
  access: open, no account

---

## 529 education savings plans

- URL: https://iinventedlight.com/library/reference/us-gov/529-plans/
- Markdown: https://iinventedlight.com/library/reference/us-gov/529-plans.md
- License: CC-BY-SA-4.0 (source: https://www.investor.gov/introduction-investing/investing-basics/glossary/529-plans)
- checked_on: 2026-10-14
- status: reviewed

# 529 education savings plans

## Summary
A 529 plan (qualified tuition program) is a tax-advantaged way to save for education, sponsored by states, state agencies or schools. Two types: prepaid tuition and education savings plans.

> **General information, not financial or tax advice.**

## Facts
- id: F1
  claim: A 529 plan, legally a qualified tuition plan, is a tax-advantaged savings vehicle for education costs, authorized by IRC Section 529.
  confidence: high
  date: 2026-10-14
  sources: [S1, S2]
  license: CC BY-SA 4.0
  checked_on: 2026-10-14
- id: F2
  claim: Plans are sponsored by states, state agencies, or educational institutions; every state and DC sponsors at least one.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
  checked_on: 2026-10-14
- id: F3
  claim: There are two types: prepaid tuition plans and education savings plans.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
  checked_on: 2026-10-14
- id: F4
  claim: Since the 2017 Tax Cuts and Jobs Act, K-12 tuition also counts as a qualified expense, alongside higher education.
  confidence: medium
  date: 2026-10-14
  sources: [S2]
  license: CC BY-SA 4.0
  checked_on: 2026-10-14
- id: F5
  claim: The SEC does not regulate or oversee 529 plans.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
  checked_on: 2026-10-14

## Pointers
- answers: Current K-12 dollar limits and qualified-expense rules
  where: https://www.irs.gov/newsroom/529-plans-questions-and-answers
  access: open

---

## Adult vaccine schedule overview (US)

- URL: https://iinventedlight.com/library/reference/us-gov/adult-vaccine-schedule/
- Markdown: https://iinventedlight.com/library/reference/us-gov/adult-vaccine-schedule.md
- License: CC-BY-4.0 (source: https://www.cdc.gov/vaccines/hcp/imz-schedules/adult-age.html)
- checked_on: 2026-10-14
- status: reviewed

# Adult vaccine schedule overview (US)

## Summary
CDC publishes an adult (19+) schedule by age and by medical condition; as of 2026-10-14 the current one is still the 2025 schedule (revised 2025-07-02), and no 2026 adult schedule was found. Routinely, adults get a flu vaccine every year and a Td/Tdap booster every 10 years; COVID-19, shingles (50+), RSV (75+, or 50–74 at increased risk), pneumococcal, and others depend on age and conditions. Check with a clinician.

> **General information, not medical advice. See a clinician about your own situation; in an emergency call 911 (US).**

## Facts
- id: F1
  claim: The adult schedule covers ages 19+ and is organized by age (Table 1) and by medical condition (Table 2), with notes, contraindications, and an addendum of newer ACIP recommendations.
  confidence: high
  date: 2026-10-14
  sources: [S1, S2]
  license: CC BY 4.0
- id: F2
  claim: Influenza: 1 dose every year; for 65+, high-dose, recombinant, or adjuvanted flu vaccine is preferred.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F3
  claim: Tdap once, then Td or Tdap every 10 years; Tdap in each pregnancy.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F4
  claim: Recombinant zoster (shingles): 2 doses at 50+, or from 19 with immunocompromising conditions.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F5
  claim: RSV: recommended at 75+, for adults 50–74 at increased risk (50–59 added June 2025), as a single dose only; seasonal dose in pregnancy.
  confidence: high
  date: 2026-10-14
  sources: [S1, S2]
  license: CC BY 4.0
- id: F6
  claim: COVID-19 (2024–2025 vaccine: 1+ doses, 2+ at 65+), pneumococcal, HPV (through 26; 27–45 by shared decision), hepatitis A/B, MMR, varicella, meningococcal, and others depend on age, history, and conditions.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F7
  claim: Since 2025-05-29 the schedule reflects an HHS directive changing COVID-19 vaccine recommendations for pregnant women.
  confidence: medium
  date: 2026-10-14
  sources: [S2]
  license: CC BY 4.0

## Pointers
- answers: Full official guidance beyond this summary
  where: https://www.cdc.gov/vaccines/hcp/imz-schedules/adult-age.html; https://www.cdc.gov/vaccines/hcp/imz-schedules/adult-addendum.html
  access: open, no account

## Open issues (pre-review)
- No 2026 adult schedule found on cdc.gov (search + pages fetched 2026-10-14). A search snippet says the schedule PDF was "amended April 27, 2026" for RSV; the PDF returned 403, so this is NOT verified and not used.
- COVID-19 row still names the 2024–2025 vaccine; recommendations may have changed since. Re-check before review.

---

## Antibiotics basics

- URL: https://iinventedlight.com/library/reference/us-gov/antibiotics-basics/
- Markdown: https://iinventedlight.com/library/reference/us-gov/antibiotics-basics.md
- License: CC-BY-4.0 (source: https://www.cdc.gov/antibiotic-use/about/index.html)
- checked_on: 2026-10-14
- status: reviewed

# Antibiotics basics

## Summary
Antibiotics treat only certain bacterial infections and do nothing for viruses like colds and flu. When prescribed, take them exactly as directed, never share or save them, and report side effects.

> **General information, not medical advice. See a clinician about your own situation; in an emergency call 911 (US).**

## Facts
- id: F1
  claim: Antibiotics treat certain bacterial infections (e.g. strep throat, whooping cough, UTIs); some bacterial infections, like many sinus and some ear infections, get better without them.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F2
  claim: They do not work on viruses: colds, flu, most sore throats, and most chest colds.
  confidence: high
  date: 2026-10-14
  sources: [S1, S2]
  license: CC BY 4.0
- id: F3
  claim: Take them exactly as prescribed; do not share them, save them for later, or take someone else's.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F4
  claim: Common side effects include rash, dizziness, nausea, diarrhea, and yeast infections; serious ones include C. diff infection, severe allergic reactions, and resistant infections.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0

## Pointers
- answers: Full official guidance and details beyond this summary
  where: https://www.cdc.gov/antibiotic-use/about/index.html; https://medlineplus.gov/commoncold.html
  access: open, no account

---

## How to check for consumer product recalls (CPSC)

- URL: https://iinventedlight.com/library/reference/us-gov/check-product-recalls/
- Markdown: https://iinventedlight.com/library/reference/us-gov/check-product-recalls.md
- License: CC-BY-4.0 (source: https://www.cpsc.gov/Recalls)
- checked_on: 2026-10-14
- status: reviewed

# How to check for consumer product recalls (CPSC)

## Summary
CPSC posts consumer product recalls and warnings at cpsc.gov/Recalls and offers recall email alerts. Most recalls offer a refund, repair, or replacement; if a recall says to stop using a product, do so right away and follow the company's instructions.

> **General information, not legal advice. Recall and warranty rights depend on the product and your state; confirm with the official source.**

## Facts
- id: F1
  claim: CPSC posts recalls and product safety warnings at cpsc.gov/Recalls; its data updates weekly as new recalls are announced.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F2
  claim: CPSC offers a subscription for recall emails.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F3
  claim: Over the past five years, CPSC recall remedies were about 54.5% refunds, 26.5% repairs, and 16.9% replacements.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
  note: Counts change daily; snapshot as of 2026-10-14. Recalls with several remedies are counted under each.
- id: F4
  claim: CPSC hazard categories include fire, burn, falls, choking, entrapment, poisoning, ingestion, and crash.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F5
  claim: CPSC also issues warnings telling consumers to stop using a product even when no recall remedy is offered.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F6
  claim: CPSC's toll-free consumer hotline is 800-638-2772 (TTY 800-638-8270).
  confidence: medium
  date: 2026-10-14
  sources: [S1, S2]
  license: CC BY 4.0
- id: F7
  claim: CPSC recall data downloads go back to 1973.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0

## Pointers
- answers: Search a specific product's recall
  where: https://www.cpsc.gov/Recalls
  access: open
- answers: Vehicles, tires and car seats (not CPSC)
  where: https://www.nhtsa.gov/recalls
  access: open

---

## FTC Cooling-Off Rule: canceling door-to-door sales

- URL: https://iinventedlight.com/library/reference/us-gov/cooling-off-rule/
- Markdown: https://iinventedlight.com/library/reference/us-gov/cooling-off-rule.md
- License: CC-BY-4.0 (source: https://consumer.ftc.gov/articles/buyers-remorse-ftcs-cooling-rule-may-help)
- checked_on: 2026-10-14
- status: reviewed

# FTC Cooling-Off Rule: canceling door-to-door sales

## Summary
The FTC Cooling-Off Rule lets you cancel certain sales made at your home, workplace, dorm, or a seller's temporary location until midnight of the third business day after the sale, for a full refund. Online, mail, and phone sales are not covered.

> **General information, not legal advice. Recall and warranty rights depend on the product and your state; confirm with the official source.**

## Facts
- id: F1
  claim: The rule covers sales at your home, workplace, or dorm, or at a seller's temporary location such as a hotel room, convention center, or fairground.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F2
  claim: You can cancel until midnight of the third business day after the sale; Saturdays count as business days, Sundays and federal holidays do not.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F3
  claim: "Not covered: home sales under $25, temporary-location sales under $130, and sales made entirely online, by mail, or by phone."
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F4
  claim: Real estate, insurance, securities, and arts or crafts sold at fairs are not covered.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F5
  claim: The seller must tell you about your right to cancel and give you two copies of a cancellation form plus a dated contract or receipt.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F6
  claim: You don't need a reason to cancel; the FTC suggests sending the form by certified mail.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F7
  claim: After you cancel, the seller has 10 days to refund your money and return any check or trade-in.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F8
  claim: Report violations to the FTC at ReportFraud.ftc.gov and to your state attorney general.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0

## Pointers
- answers: Full rule text
  where: 16 CFR Part 429 (eCFR)
  access: open

---

## CPR basics

- URL: https://iinventedlight.com/library/reference/us-gov/cpr-basics/
- Markdown: https://iinventedlight.com/library/reference/us-gov/cpr-basics.md
- License: CC-BY-SA-4.0 (source: https://medlineplus.gov/ency/article/000010.htm)
- checked_on: 2026-10-14
- status: reviewed

# CPR basics

## Summary
CPR combines chest compressions and (often) rescue breaths to keep oxygenated blood flowing when breathing or heartbeat stops. Call 911 first; the dispatcher can coach you. Take a certified class for technique.

> **General information, not medical advice. See a clinician about your own situation; in an emergency call 911 (US).**

## Facts
- id: F1
  claim: CPR is an emergency procedure for when a person's breathing or heartbeat has stopped (cardiac or respiratory arrest).
  confidence: high
  date: 2026-10-14
  sources: [S1, S2]
  license: CC BY-SA 4.0
  checked_on: 2026-10-14
- id: F2
  claim: CPR uses chest compressions, often combined with rescue breathing, to keep blood and oxygen circulating until normal function is restored or help arrives.
  confidence: high
  date: 2026-10-14
  sources: [S1, S2]
  license: CC BY-SA 4.0
  checked_on: 2026-10-14
- id: F3
  claim: Brain damage or death can occur within about 4 minutes once blood flow stops.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
  checked_on: 2026-10-14
- id: F4
  claim: 911 operators can guide a bystander through CPR.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
  checked_on: 2026-10-14
- id: F5
  claim: Technique differs for adults/teens, children age 1 to puberty, and infants under 1.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
  checked_on: 2026-10-14

## Pointers
- answers: Step-by-step compression depth/rate and hands-only CPR training
  where: https://cpr.heart.org (AHA 2025 Guidelines)
  access: open site; content all-rights-reserved, pointer only

---

## What a credit score is

- URL: https://iinventedlight.com/library/reference/us-gov/credit-scores/
- Markdown: https://iinventedlight.com/library/reference/us-gov/credit-scores.md
- License: CC-BY-4.0 (source: https://www.consumerfinance.gov/ask-cfpb/what-is-a-credit-score-en-315/)
- checked_on: 2026-10-14
- status: reviewed

# What a credit score is

## Summary
A credit score predicts how likely you are to repay credit, computed by a scoring model from your credit reports. You have many scores, not one; most range 300-850.

> **General information, not financial or tax advice. Check IRS/agency sources or a professional for your situation.**

## Facts
- id: F1
  claim: A credit score predicts credit behavior, such as likelihood of repaying a loan on time, based on credit report data.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
  checked_on: 2026-10-14
- id: F2
  claim: Lenders, landlords and insurers use scores, which also affect interest rates and credit limits.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
  checked_on: 2026-10-14
- id: F3
  claim: Common factors: payment history, current debt, number/type of accounts, account age, credit utilization, new applications, collections/foreclosure/bankruptcy.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
  checked_on: 2026-10-14
- id: F4
  claim: You have multiple scores; they vary by model, data source and date.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
  checked_on: 2026-10-14
- id: F5
  claim: Most credit scores range from 300 to 850; higher generally means better terms.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
  checked_on: 2026-10-14

## Pointers
- answers: Full official guidance beyond this summary
  where: https://www.consumerfinance.gov/ask-cfpb/what-is-a-credit-score-en-315/
  access: open, no account

---

## Depression basics

- URL: https://iinventedlight.com/library/reference/us-gov/depression-basics/
- Markdown: https://iinventedlight.com/library/reference/us-gov/depression-basics.md
- License: CC-BY-4.0 (source: https://www.nimh.nih.gov/health/topics/depression)
- checked_on: 2026-10-14
- status: reviewed

# Depression basics

## Summary
Depression (major depressive disorder) is a treatable illness, different from ordinary sadness, with symptoms that last at least two weeks and affect daily life. Treatment usually involves medicine and/or talk therapy. In crisis in the US, call or text 988.

> **General information, not medical advice. See a clinician about your own situation; in an emergency call 911 (US).**

## Facts
- id: F1
  claim: Depression differs from passing sadness: it causes symptoms that affect feeling, thinking, and daily activities such as sleeping, eating, or working, and can affect anyone.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F2
  claim: Genetic, biological, environmental, and psychological factors are thought to play a role.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F3
  claim: Symptoms can include persistent sadness or emptiness, lost interest in activities, appetite and sleep changes, fatigue, hopelessness or guilt, aches, and thoughts of death or suicide; they can look different in children, teens, men, and older adults.
  confidence: medium
  date: 2026-10-14
  sources: [S2]
  license: CC BY 4.0
- id: F4
  claim: Diagnosis requires symptoms most of the day, nearly every day, for at least two weeks, including depressed mood or loss of interest.
  confidence: medium
  date: 2026-10-14
  sources: [S2]
  license: CC BY 4.0
- id: F5
  claim: Most treatment includes medicines and/or psychotherapy; antidepressants can take time to work and should not be stopped without talking to a provider.
  confidence: medium
  date: 2026-10-14
  sources: [S2]
  license: CC BY 4.0
- id: F6
  claim: In the US, someone thinking about harming themselves can call or text 988 (988 Suicide & Crisis Lifeline).
  confidence: medium
  date: 2026-10-14
  sources: [S3]
  license: CC BY 4.0

## Pointers
- answers: Full official guidance and details beyond this summary
  where: https://www.nimh.nih.gov/health/topics/depression; https://medlineplus.gov/depression.html; https://medlineplus.gov/ency/article/001927.htm
  access: open, no account

---

## Disputing a credit card charge

- URL: https://iinventedlight.com/library/reference/us-gov/dispute-credit-card-charge/
- Markdown: https://iinventedlight.com/library/reference/us-gov/dispute-credit-card-charge.md
- License: CC-BY-4.0 (source: https://consumer.ftc.gov/articles/using-credit-cards-and-disputing-charges)
- checked_on: 2026-10-14
- status: reviewed

# Disputing a credit card charge

## Summary
For billing errors, write to the card issuer's billing-inquiries address so it arrives within 60 days of the first bill showing the error. The issuer must acknowledge within 30 days and resolve within 90; you may withhold the disputed amount meanwhile.

> **General information, not legal advice.**

## Facts
- id: F1
  claim: The Fair Credit Billing Act sets a dispute process for billing errors on credit cards and other revolving credit, not installment loans.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
  checked_on: 2026-10-14
- id: F2
  claim: Disputable errors include unauthorized charges, wrong amounts or dates, and items not accepted or not delivered as agreed.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
  checked_on: 2026-10-14
- id: F3
  claim: Liability for unauthorized credit card charges is capped at $50 under federal law.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
  checked_on: 2026-10-14
- id: F4
  claim: Send a written dispute to the billing-inquiries address (not the payment address) so it arrives within 60 days after the first bill with the error.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
  checked_on: 2026-10-14
- id: F5
  claim: The issuer must acknowledge in writing within 30 days and resolve within 90 days.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
  checked_on: 2026-10-14
- id: F6
  claim: During the investigation you may withhold the disputed amount but must pay the undisputed portion; the issuer can't report it as delinquent.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
  checked_on: 2026-10-14
- id: F7
  claim: For poor-quality purchases, you generally must first try the seller; the federal right applies to purchases over $5 made in-state or within 100 miles of your billing address.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
  checked_on: 2026-10-14

## Pointers
- answers: Full official guidance beyond this summary
  where: https://consumer.ftc.gov/articles/using-credit-cards-and-disputing-charges
  access: open, no account

---

## Earthquake magnitude vs. intensity

- URL: https://iinventedlight.com/library/reference/us-gov/earthquake-magnitude/
- Markdown: https://iinventedlight.com/library/reference/us-gov/earthquake-magnitude.md
- License: CC-BY-4.0 (source: https://www.usgs.gov/programs/earthquake-hazards/earthquake-magnitude-energy-release-and-shaking-intensity)
- checked_on: 2026-10-14
- status: reviewed

# Earthquake magnitude vs. intensity

## Summary
Magnitude is one number for an earthquake's size; each whole-number step means 10x the measured wave amplitude and about 32x the energy. Intensity (Modified Mercalli, Roman numerals) is the shaking at a given place and varies by distance, depth, and local ground. Moment magnitude (Mw) has largely replaced the Richter scale.

> **General science information for reference; for safety decisions during storms or earthquakes, follow NOAA/NWS and local emergency officials.**

## Facts
- id: F1
  claim: An earthquake has a single magnitude, but shaking intensity varies from place to place.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F2
  claim: Each whole-number increase in magnitude is a tenfold increase in measured amplitude on a seismogram.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F3
  claim: Each whole-number increase in magnitude releases about 32 times more energy.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F4
  claim: The Richter scale (ML), introduced by Charles Richter in 1935, is now mostly used only for small local earthquakes; moment magnitude (Mw) is more accurate for others.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F5
  claim: Moment magnitude is based on seismic moment, which is rock rigidity times fault area times slip.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F6
  claim: The U.S. uses the Modified Mercalli Intensity scale, written in Roman numerals.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F7
  claim: Shallower earthquakes and soft ground generally produce stronger shaking than deep ones or bedrock.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F8
  claim: USGS considers a magnitude 10.5 earthquake highly unlikely because Earth lacks a long enough fault system.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0

## Pointers
- answers: Magnitude types table
  where: https://www.usgs.gov/programs/earthquake-hazards/earthquake-magnitude-energy-release-and-shaking-intensity
  access: open

---

## What the ENERGY STAR label means

- URL: https://iinventedlight.com/library/reference/us-gov/energy-star/
- Markdown: https://iinventedlight.com/library/reference/us-gov/energy-star.md
- License: CC-BY-SA-4.0 (source: https://www.energystar.gov/about)
- checked_on: 2026-10-14
- status: reviewed

# What the ENERGY STAR label means

## Summary
ENERGY STAR is the U.S. government-backed energy-efficiency label, run by the EPA since 1992; it's a voluntary program for products and buildings.

> **General information, not legal advice.**

## Facts
- id: F1
  claim: ENERGY STAR is a government-backed energy-efficiency symbol administered by the U.S. EPA.
  confidence: high
  date: 2026-10-14
  sources: [S1, S2]
  license: CC BY-SA 4.0
  checked_on: 2026-10-14
- id: F2
  claim: The program began in 1992.
  confidence: high
  date: 2026-10-14
  sources: [S1, S2]
  license: CC BY-SA 4.0
  checked_on: 2026-10-14
- id: F3
  claim: Participation is voluntary; thousands of organizations partner with the program.
  confidence: high
  date: 2026-10-14
  sources: [S1, S2]
  license: CC BY-SA 4.0
  checked_on: 2026-10-14
- id: F4
  claim: EPA reports the program has helped save 5 trillion kWh and avoid more than $500 billion in energy costs since 1992.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
  checked_on: 2026-10-14

## Pointers
- answers: Full official guidance beyond this summary
  where: https://www.energystar.gov/about; https://en.wikipedia.org/wiki/Energy_Star
  access: open, no account

---

## Main types of federal student aid

- URL: https://iinventedlight.com/library/reference/us-gov/fafsa-aid-types/
- Markdown: https://iinventedlight.com/library/reference/us-gov/fafsa-aid-types.md
- License: CC-BY-SA-4.0 (source: https://en.wikipedia.org/wiki/FAFSA)
- checked_on: 2026-10-14
- status: reviewed

# Main types of federal student aid

## Summary
Federal student aid comes as grants (such as Pell, which generally isn't repaid), work-study, and loans. Direct Subsidized Loans are need-based and the government pays interest while you're in school at least half time; Direct Unsubsidized Loans are not need-based and accrue interest from the start.

> **General information, not financial, legal, or academic advice. Aid rules and amounts change each award year; confirm at StudentAid.gov or with your school's financial aid office.**

## Facts
- id: F1
  claim: Federal Pell Grants are need-based grants that generally do not have to be repaid.
  confidence: high
  date: 2026-10-14
  sources: [S1, S2]
  license: CC BY-SA 4.0
- id: F3
  claim: Direct Subsidized Loans are need-based; the government pays the interest while the student is enrolled at least half time.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
- id: F4
  claim: Direct Unsubsidized Loans are not need-based and accrue interest from disbursement.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
- id: F5
  claim: Federal Work-Study subsidizes part-time jobs for students with financial need.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
- id: F6
  claim: Graduate and professional students are not eligible for Direct Subsidized Loans.
  confidence: medium
  date: 2026-10-14
  sources: [S3]
  license: CC BY-SA 4.0

## Pointers
- answers: Official descriptions of each aid type
  where: https://studentaid.gov/understand-aid/types
  access: open (JavaScript app; could not be fetched automatically on 2026-10-14, so not used as a source)

---

## FAFSA: when it opens, deadlines, and cost

- URL: https://iinventedlight.com/library/reference/us-gov/fafsa-deadlines/
- Markdown: https://iinventedlight.com/library/reference/us-gov/fafsa-deadlines.md
- License: CC-BY-SA-4.0 (source: https://studentaid.gov/sites/default/files/2027-28-fafsa-form.pdf)
- checked_on: 2026-10-14
- status: reviewed

# FAFSA: when it opens, deadlines, and cost

## Summary
For 2027–28 aid, the FAFSA can be submitted from October 1, 2026, and federal must receive it by June 30, 2028; state and school deadlines can be much earlier, so file as soon as possible. The FAFSA is free; the official site is fafsa.gov / StudentAid.gov.

> **General information, not financial, legal, or academic advice. Aid rules and amounts change each award year; confirm at StudentAid.gov or with your school's financial aid office.**

## Facts
- id: F1
  claim: The 2027–28 FAFSA covers school attendance from July 1, 2027 through June 30, 2028.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
- id: F2
  claim: For federal aid on the 2027–28 FAFSA, apply no earlier than October 1, 2026; the Department must receive the application by June 30, 2028.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
- id: F3
  claim: State or college aid deadlines can be as early as October 1, 2026, or that aid may be first-come, first-served.
  confidence: high
  date: 2026-10-14
  sources: [S1, S2]
  license: CC BY-SA 4.0
- id: F4
  claim: Washington State lists its deadline as as soon as possible after October 1, 2026, with funds available year round.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
  note: Other states differ; see page 2 of the form.
- id: F5
  claim: The FAFSA form itself is free; it can be filed online at fafsa.gov.
  confidence: high
  date: 2026-10-14
  sources: [S1, S2]
  license: CC BY-SA 4.0
  note: Wikipedia warns some look-alike sites charge fees.
- id: F6
  claim: Federal law (FAFSA Deadline Act, 2024) requires the FAFSA to be available by October 1 each year.
  confidence: medium
  date: 2026-10-14
  sources: [S2]
  license: CC BY-SA 4.0
  note: Federal Student Aid also referred to launching the 2026–27 form 'by Oct. 1, 2025' [S3].
- id: F7
  claim: The FAFSA uses tax information from two years earlier ('prior-prior year').
  confidence: medium
  date: 2026-10-14
  sources: [S2]
  license: CC BY-SA 4.0
- id: F8
  claim: A paper FAFSA can list up to ten colleges.
  confidence: high
  date: 2026-10-14
  sources: [S1, S2]
  license: CC BY-SA 4.0

## Pointers
- answers: Every state's 2027–28 FAFSA deadline
  where: https://studentaid.gov/sites/default/files/2027-28-fafsa-form.pdf (page 2)
  access: open, no account
- answers: File the FAFSA online
  where: https://fafsa.gov
  access: free StudentAid.gov account (FSA ID) required

---

## Fastest-growing occupations (BLS projections 2025–35)

- URL: https://iinventedlight.com/library/reference/us-gov/fastest-growing-occupations/
- Markdown: https://iinventedlight.com/library/reference/us-gov/fastest-growing-occupations.md
- License: CC-BY-4.0 (source: https://www.bls.gov/ooh/fastest-growing.htm)
- checked_on: 2026-10-14
- status: reviewed

# Fastest-growing occupations (BLS projections 2025–35)

## Summary
BLS projects nurse practitioners (41%), solar photovoltaic installers (37%), and data scientists (35%) as the fastest-growing occupations for 2025–35, followed by wind turbine technicians (30%) and medical and health services managers (24%). Health care dominates the top 20.

> **General information, not legal advice. State laws may give more protection; check your state labor department.**

## Facts
- id: F1
  claim: Nurse practitioners have the highest projected growth, 41% for 2025–35, with 2025 median pay of $132,300.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F2
  claim: Solar photovoltaic installers are projected to grow 37% (median $53,140).
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F3
  claim: Data scientists are projected to grow 35% (median $120,230).
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F4
  claim: Wind turbine service technicians are projected to grow 30%; medical and health services managers 24%.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F5
  claim: Information security analysts (21%, median $129,180) and computer and information research scientists (22%) are also in the top 20.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F6
  claim: The BLS page was last modified August 27, 2026.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0

## Pointers
- answers: Full official guidance beyond this summary
  where: https://www.bls.gov/ooh/fastest-growing.htm
  access: open, no account

---

## FDIC deposit insurance

- URL: https://iinventedlight.com/library/reference/us-gov/fdic-deposit-insurance/
- Markdown: https://iinventedlight.com/library/reference/us-gov/fdic-deposit-insurance.md
- License: CC-BY-4.0 (source: https://www.fdic.gov/resources/deposit-insurance/understanding-deposit-insurance)
- checked_on: 2026-10-14
- status: reviewed

# FDIC deposit insurance

## Summary
FDIC insurance covers deposits up to $250,000 per depositor, per FDIC-insured bank, per ownership category. It covers checking, savings, money market deposit accounts, and CDs, not stocks, bonds, mutual funds, annuities, crypto, or safe deposit boxes.

> **General information, not financial or tax advice. Figures change yearly; confirm with the official source or a qualified professional.**

## Facts
- id: F1
  claim: Coverage is $250,000 per depositor, per FDIC-insured bank, for each account ownership category.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F2
  claim: Covered: checking, NOW, savings, money market deposit accounts, CDs, and bank-issued cashier's checks and money orders.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F3
  claim: Not covered: stocks, bonds, mutual funds, annuities, life insurance, safe deposit box contents, Treasury securities, municipal securities, and crypto assets.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F4
  claim: Ownership categories include single, joint, certain retirement accounts such as IRAs, trust, and business accounts; each is insured separately.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F5
  claim: The FDIC says no depositor has lost FDIC-insured funds since it was founded in 1933.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F6
  claim: You can check whether a bank is FDIC-insured with the FDIC BankFind Suite tool.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0

## Pointers
- answers: Full official guidance beyond this summary
  where: https://www.fdic.gov/resources/deposit-insurance/understanding-deposit-insurance
  access: open, no account

---

## US federal minimum wage

- URL: https://iinventedlight.com/library/reference/us-gov/federal-minimum-wage/
- Markdown: https://iinventedlight.com/library/reference/us-gov/federal-minimum-wage.md
- License: CC-BY-4.0 (source: https://www.dol.gov/agencies/whd/minimum-wage)
- checked_on: 2026-10-14
- status: reviewed

# US federal minimum wage

## Summary
The federal minimum wage is $7.25 per hour, in effect since July 24, 2009, under the Fair Labor Standards Act. Many states set higher minimums; employers must follow both.

> **General information, not legal advice. State laws may give more protection; check your state labor department.**

## Facts
- id: F1
  claim: The federal minimum wage is $7.25 per hour, effective July 24, 2009.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F2
  claim: Federal minimum wage rules are in the Fair Labor Standards Act (FLSA).
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F3
  claim: Many states have their own minimum wage laws; when both apply, employers must comply with both.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F4
  claim: The FLSA does not provide a way to collect promised wages above its required minimums, though some state laws do.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F5
  claim: Wage and Hour Division help line: 1-866-487-9243.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0

## Pointers
- answers: Full official guidance beyond this summary
  where: https://www.dol.gov/agencies/whd/minimum-wage
  access: open, no account

---

## Federal income tax filing deadline

- URL: https://iinventedlight.com/library/reference/us-gov/federal-tax-filing-deadline/
- Markdown: https://iinventedlight.com/library/reference/us-gov/federal-tax-filing-deadline.md
- License: CC-BY-4.0 (source: https://www.irs.gov/filing/individuals/when-to-file)
- checked_on: 2026-10-14
- status: reviewed

# Federal income tax filing deadline

## Summary
Calendar-year individual returns were due April 15, 2026 (moving to the next business day if it falls on a weekend or holiday). Form 4868 gives an automatic 6-month extension to file, but not to pay.

> **General information, not financial or tax advice. Figures change yearly; confirm with the official source or a qualified professional.**

## Facts
- id: F1
  claim: For calendar-year filers the IRS lists the 2026 filing date as April 15, 2026.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F2
  claim: If the 15th falls on a Saturday, Sunday, or legal holiday, the deadline moves to the next business day.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F3
  claim: A mailed return is on time if properly addressed, postmarked, and mailed by the due date.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F4
  claim: Form 4868, filed by the original due date, gives an automatic 6-month extension to file.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F5
  claim: An extension to file is not an extension to pay; tax owed should be paid by the original date to avoid penalties.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F6
  claim: Fiscal-year filers file by the 15th day of the fourth month after their fiscal year ends.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0

## Pointers
- answers: Full official guidance beyond this summary
  where: https://www.irs.gov/filing/individuals/when-to-file
  access: open, no account

---

## How to file for unemployment insurance

- URL: https://iinventedlight.com/library/reference/us-gov/file-for-unemployment/
- Markdown: https://iinventedlight.com/library/reference/us-gov/file-for-unemployment.md
- License: CC-BY-4.0 (source: https://www.dol.gov/general/topic/unemployment-insurance)
- checked_on: 2026-10-14
- status: reviewed

# How to file for unemployment insurance

## Summary
Unemployment insurance is a joint state-federal program run by each state. File with the state where you worked, as soon as possible after losing your job; the first payment usually takes 2–3 weeks.

> **General information, not legal advice. State laws may give more protection; check your state labor department.**

## Facts
- id: F1
  claim: Each state runs its own unemployment insurance program under federal guidelines.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F2
  claim: You usually qualify if you lost your job through no fault of your own and meet your state's wage or work requirements for a "base period".
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F3
  claim: In most states the base period is the first four of the last five completed calendar quarters.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F4
  claim: File with the state where you worked; depending on the state, claims are filed online, by phone, or in person.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F5
  claim: The first benefit payment generally arrives 2 to 3 weeks after filing.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F6
  claim: DOL's toll-free line for job loss and benefits questions: 1-877-US-2JOBS (TTY 1-877-889-5627).
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0

## Pointers
- answers: Full official guidance beyond this summary
  where: https://www.dol.gov/general/topic/unemployment-insurance
  access: open, no account

---

## Flu vs. cold vs. COVID-19

- URL: https://iinventedlight.com/library/reference/us-gov/flu-cold-covid/
- Markdown: https://iinventedlight.com/library/reference/us-gov/flu-cold-covid.md
- License: CC-BY-4.0 (source: https://medlineplus.gov/flu.html)
- checked_on: 2026-10-14
- status: reviewed

# Flu vs. cold vs. COVID-19

## Summary
All three are viral respiratory infections. Colds usually start gradually and stay mild; flu tends to start suddenly with fever, aches, and fatigue; flu and COVID-19 share many symptoms and cannot be told apart without a test. COVID-19 spreads more easily, can take longer to show symptoms, and stays contagious longer. Antibiotics treat none of them.

> **General information, not medical advice. See a clinician about your own situation; in an emergency call 911 (US).**

## Facts
- id: F1
  claim: Colds are mild upper-respiratory infections caused by 200+ viruses, most often rhinoviruses; adults average 2–3 a year.
  confidence: medium
  date: 2026-10-14
  sources: [S2]
  license: CC BY 4.0
- id: F2
  claim: Compared with a cold, flu starts suddenly and usually brings fever, aches, and fatigue; colds start slowly and more often cause stuffy nose, sneezing, or sore throat.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F3
  claim: Flu is caused by influenza viruses; COVID-19 by the coronavirus SARS-CoV-2.
  confidence: high
  date: 2026-10-14
  sources: [S3, S4]
  license: CC BY 4.0
- id: F4
  claim: Flu and COVID-19 share symptoms (fever/chills, cough, shortness of breath, fatigue, sore throat, runny nose, aches, headache, vomiting, diarrhea); loss of taste or smell is more frequent with COVID-19. Symptoms alone cannot tell them apart; testing is needed.
  confidence: high
  date: 2026-10-14
  sources: [S3, S4]
  license: CC BY 4.0
- id: F5
  claim: Flu symptoms typically appear 1–4 days after infection; COVID-19 symptoms 2–5 days and up to 14 days.
  confidence: high
  date: 2026-10-14
  sources: [S3, S4]
  license: CC BY 4.0
  note: CDC symptoms page says 2–14 days.
- id: F6
  claim: COVID-19 is generally more contagious than flu and people may stay contagious longer (about 8 days after symptoms start on average; spread can begin 2–3 days before symptoms).
  confidence: medium
  date: 2026-10-14
  sources: [S3]
  license: CC BY 4.0
- id: F7
  claim: Flu tends to cause more severe illness in young children than COVID-19; COVID-19 can also lead to blood clots and Long COVID.
  confidence: medium
  date: 2026-10-14
  sources: [S3]
  license: CC BY 4.0
- id: F8
  claim: Antivirals: flu drugs work best within 2 days of getting sick; COVID-19 antivirals for higher-risk people must start within 5–7 days of symptoms. Flu antivirals do not treat COVID-19.
  confidence: high
  date: 2026-10-14
  sources: [S1, S3]
  license: CC BY 4.0
- id: F9
  claim: There is no cure or vaccine for the common cold; rest and fluids help, and most people feel better in 1–2 weeks.
  confidence: medium
  date: 2026-10-14
  sources: [S2]
  license: CC BY 4.0
- id: F10
  claim: Antibiotics do not work on viruses such as those causing colds and flu.
  confidence: high
  date: 2026-10-14
  sources: [S5, S2]
  license: CC BY 4.0

## Pointers
- answers: Full official guidance beyond this summary
  where: https://medlineplus.gov/flu.html; https://medlineplus.gov/commoncold.html; https://www.cdc.gov/flu/about/flu-vs-covid19.html; https://www.cdc.gov/covid/signs-symptoms/index.html; https://www.cdc.gov/antibiotic-use/about/index.html
  access: open, no account

## Open issues (pre-review)
- CDC flu-vs-COVID page is dated 2024-09-17; vaccine statements there were not used.

---

## Family and Medical Leave Act (FMLA)

- URL: https://iinventedlight.com/library/reference/us-gov/fmla-leave/
- Markdown: https://iinventedlight.com/library/reference/us-gov/fmla-leave.md
- License: CC-BY-4.0 (source: https://www.dol.gov/agencies/whd/fmla)
- checked_on: 2026-10-14
- status: reviewed

# Family and Medical Leave Act (FMLA)

## Summary
FMLA gives eligible employees up to 12 workweeks of unpaid, job-protected leave in 12 months for qualifying family and medical reasons (26 weeks for military caregiver leave), with group health coverage continued. Eligibility: 12 months with the employer, 1,250 hours in the past year, and 50+ employees within 75 miles.

> **General information, not legal advice. State laws may give more protection; check your state labor department.**

## Facts
- id: F1
  claim: FMLA leave is unpaid and job-protected, and group health benefits continue on the same terms.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F2
  claim: Eligible employees get up to 12 workweeks in a 12-month period for a birth, an adoption or foster placement, a serious health condition of the employee or a spouse, child, or parent, or qualifying military exigencies.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F3
  claim: Military caregiver leave allows up to 26 workweeks in a single 12-month period.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F4
  claim: Covered employers: private employers with 50+ employees in 20+ workweeks this or last year, all public agencies, and all public and private elementary and secondary schools.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F5
  claim: Employees qualify after 12 months with the employer, 1,250 hours in the prior 12 months, at a site with 50+ employees within 75 miles.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F6
  claim: Leave can be intermittent or on a reduced schedule when medically necessary; for bonding with a new child, only if the employer agrees.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0

## Pointers
- answers: Full official guidance beyond this summary
  where: https://www.dol.gov/agencies/whd/fmla
  access: open, no account

---

## How to make a FOIA request

- URL: https://iinventedlight.com/library/reference/us-gov/foia-request/
- Markdown: https://iinventedlight.com/library/reference/us-gov/foia-request.md
- License: CC-BY-4.0 (source: https://www.foia.gov/how-to.html)
- checked_on: 2026-10-14
- status: reviewed

# How to make a FOIA request

## Summary
First check whether the record is already public. Otherwise send a written request to the agency's FOIA office that reasonably describes the records; no special form is required and most agencies accept electronic requests.

> **General information, not legal advice. Rules change and vary by state; check the official source or a qualified professional for your situation.**

## Facts
- id: F1
  claim: Check first whether the information is already public on the agency website or FOIA.gov.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F2
  claim: A FOIA request must be in writing and reasonably describe the records sought.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F3
  claim: There is no required form; most agencies take requests by web form, email, or fax.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F4
  claim: You can request any agency record and say what format you want it in.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F5
  claim: Agencies do not have to create new records, do research, analyze data, or answer questions.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F6
  claim: Response time depends on complexity and backlog; narrower requests are usually processed faster.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0

## Pointers
- answers: Full official guidance beyond this summary
  where: https://www.foia.gov/how-to.html
  access: open, no account

---

## Free credit reports

- URL: https://iinventedlight.com/library/reference/us-gov/free-credit-reports/
- Markdown: https://iinventedlight.com/library/reference/us-gov/free-credit-reports.md
- License: CC-BY-4.0 (source: https://consumer.ftc.gov/articles/free-credit-reports)
- checked_on: 2026-10-14
- status: reviewed

# Free credit reports

## Summary
Get free credit reports from Equifax, Experian, and TransUnion at AnnualCreditReport.com, the only authorized site; weekly free reports are now permanent. Check all three and dispute errors with the bureau and the business that supplied the data.

> **General information, not financial or tax advice. Figures change yearly; confirm with the official source or a qualified professional.**

## Facts
- id: F1
  claim: AnnualCreditReport.com (or 1-877-322-8228) is the only authorized source for free annual credit reports.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F2
  claim: The three nationwide bureaus permanently let consumers check each report once a week for free.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F3
  claim: Federal law (FCRA) also requires each bureau to give a free report every 12 months on request.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F4
  claim: Review all three reports, since each bureau may have different information.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F5
  claim: Dispute errors with the credit bureau and with the business that supplied the information.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F6
  claim: Online orders are available immediately; phone or mail orders are processed and mailed within 15 days.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0

## Pointers
- answers: Full official guidance beyond this summary
  where: https://consumer.ftc.gov/articles/free-credit-reports
  access: open, no account

---

## High blood pressure (hypertension)

- URL: https://iinventedlight.com/library/reference/us-gov/high-blood-pressure/
- Markdown: https://iinventedlight.com/library/reference/us-gov/high-blood-pressure.md
- License: CC-BY-4.0 (source: https://medlineplus.gov/highbloodpressure.html)
- checked_on: 2026-10-14
- status: reviewed

# High blood pressure (hypertension)

## Summary
High blood pressure usually causes no symptoms, so regular measurement is the only way to know. CDC (citing the 2025 ACC/AHA guideline) defines it as readings consistently at or above 130/80 mm Hg; treatment is lifestyle change and, if needed, medicine.

> **General information, not medical advice. See a clinician about your own situation; in an emergency call 911 (US).**

## Facts
- id: F1
  claim: Blood pressure is the force of blood against artery walls; the top (systolic) number is pressure during a heartbeat, the bottom (diastolic) between beats.
  confidence: high
  date: 2026-10-14
  sources: [S1, S2]
  license: CC BY 4.0
- id: F2
  claim: Normal is below 120/80 mm Hg; hypertension is consistently 130/80 or higher; "elevated" is systolic 120–129 with diastolic under 80.
  confidence: medium
  date: 2026-10-14
  sources: [S2]
  license: CC BY 4.0
  note: Thresholds come from the 2025 ACC/AHA guideline as presented by CDC; MedlinePlus summary page does not state numeric cutoffs.
- id: F3
  claim: It usually has no warning signs; checking blood pressure is the only way to find out.
  confidence: high
  date: 2026-10-14
  sources: [S1, S2]
  license: CC BY 4.0
- id: F4
  claim: Untreated, it raises the risk of heart attack, stroke, heart failure, and kidney disease.
  confidence: high
  date: 2026-10-14
  sources: [S1, S2]
  license: CC BY 4.0
- id: F5
  claim: Treatment includes heart-healthy lifestyle changes (activity, not smoking, diet with less sodium and alcohol, healthy weight, stress management) and medicine when changes are not enough.
  confidence: high
  date: 2026-10-14
  sources: [S1, S2]
  license: CC BY 4.0

## Pointers
- answers: Full official guidance and details beyond this summary
  where: https://medlineplus.gov/highbloodpressure.html; https://www.cdc.gov/high-blood-pressure/about/index.html
  access: open, no account

---

## Hurricane categories (Saffir-Simpson scale)

- URL: https://iinventedlight.com/library/reference/us-gov/hurricane-categories/
- Markdown: https://iinventedlight.com/library/reference/us-gov/hurricane-categories.md
- License: CC-BY-4.0 (source: https://www.nhc.noaa.gov/aboutsshws.php)
- checked_on: 2026-10-14
- status: reviewed

# Hurricane categories (Saffir-Simpson scale)

## Summary
The Saffir-Simpson scale rates hurricanes 1–5 by maximum sustained wind only: Cat 1 74–95 mph, Cat 2 96–110, Cat 3 111–129, Cat 4 130–156, Cat 5 157+. Category 3 and above are 'major' hurricanes. It ignores storm surge, flooding, and tornadoes.

> **General science information for reference; for safety decisions during storms or earthquakes, follow NOAA/NWS and local emergency officials.**

## Facts
- id: F1
  claim: The Saffir-Simpson Hurricane Wind Scale rates hurricanes 1 to 5 based only on maximum sustained wind speed.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F2
  claim: The scale does not account for storm surge, rainfall flooding, or tornadoes.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F3
  claim: Category 1 is 74–95 mph (119–153 km/h).
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F4
  claim: Category 2 is 96–110 mph (154–177 km/h).
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F5
  claim: Category 3 is 111–129 mph (178–208 km/h).
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F6
  claim: Category 4 is 130–156 mph (209–251 km/h).
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F7
  claim: Category 5 is 157 mph (252 km/h) or higher.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F8
  claim: Category 3 and higher are called major hurricanes.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F9
  claim: In the western North Pacific, 'super typhoon' means sustained winds over 150 mph.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0

## Pointers
- answers: Damage descriptions per category
  where: https://www.nhc.noaa.gov/aboutsshws.php
  access: open

---

## Hydration

- URL: https://iinventedlight.com/library/reference/us-gov/hydration/
- Markdown: https://iinventedlight.com/library/reference/us-gov/hydration.md
- License: CC-BY-4.0 (source: https://medlineplus.gov/dehydration.html)
- checked_on: 2026-10-14
- status: reviewed

# Hydration

## Summary
Fluid needs vary by age, sex, activity, pregnancy, and breastfeeding, so there is no single daily number; water and other drinks plus water-rich foods count. Drink more in heat, exercise, fever, or vomiting/diarrhea. Thirst, dry mouth, dark urine, and dizziness signal dehydration; confusion or fainting needs urgent care.

> **General information, not medical advice. See a clinician about your own situation; in an emergency call 911 (US).**

## Facts
- id: F1
  claim: Dehydration means losing more fluid than you take in; causes include diarrhea, vomiting, heavy sweating, and frequent urination.
  confidence: high
  date: 2026-10-14
  sources: [S1, S2]
  license: CC BY 4.0
- id: F2
  claim: Adult signs include strong thirst, dry mouth, less urination and sweating, dark urine, dry skin, tiredness, and dizziness.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F3
  claim: Water helps keep normal body temperature, cushion joints, protect the spinal cord, and remove waste.
  confidence: medium
  date: 2026-10-14
  sources: [S2]
  license: CC BY 4.0
- id: F4
  claim: Daily water needs vary by age, sex, pregnancy, activity, and breastfeeding; ask a provider for a personal amount.
  confidence: high
  date: 2026-10-14
  sources: [S1, S2]
  license: CC BY 4.0
- id: F5
  claim: Most intake comes from water and other drinks; water-rich foods like many fruits and vegetables also count.
  confidence: medium
  date: 2026-10-14
  sources: [S2]
  license: CC BY 4.0
- id: F6
  claim: You need more fluid in hot weather, with more activity, with fever, and with diarrhea or vomiting.
  confidence: high
  date: 2026-10-14
  sources: [S1, S2]
  license: CC BY 4.0
- id: F7
  claim: Higher-risk groups include older adults (thirst can fade with age), infants and young children, some people with chronic illness, and people active outdoors in heat.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F8
  claim: Mild cases: drink water (sports drinks if electrolytes were lost; watch the sugar); severe cases may need IV fluids.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F9
  claim: Get medical help right away for confusion, fainting, no urination, or rapid heartbeat or breathing.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0

## Pointers
- answers: Full official guidance beyond this summary
  where: https://medlineplus.gov/dehydration.html; https://www.cdc.gov/healthy-weight-growth/water-healthy-drinks/index.html
  access: open, no account

---

## Ibuprofen vs. acetaminophen

- URL: https://iinventedlight.com/library/reference/us-gov/ibuprofen-vs-acetaminophen/
- Markdown: https://iinventedlight.com/library/reference/us-gov/ibuprofen-vs-acetaminophen.md
- License: CC-BY-4.0 (source: https://medlineplus.gov/druginfo/meds/a682159.html)
- checked_on: 2026-10-14
- status: reviewed

# Ibuprofen vs. acetaminophen

## Summary
Both are over-the-counter pain relievers/fever reducers. Ibuprofen is an NSAID with stomach-bleeding and heart-attack/stroke warnings; acetaminophen is not an NSAID, and its main danger is severe liver damage from too much (over 4,000 mg in 24 hours), often from doubling up on products that contain it.

> **General information, not medical advice. See a clinician about your own situation; in an emergency call 911 (US).**

## Facts
- id: F1
  claim: Both are used to temporarily relieve minor aches and pains (headache, toothache, backache, menstrual cramps, colds, muscle aches, minor arthritis pain) and to reduce fever.
  confidence: high
  date: 2026-10-14
  sources: [S1, S2, S3, S4]
  license: CC BY 4.0
- id: F2
  claim: Ibuprofen is an NSAID; MedlinePlus says it reduces pain, fever, and inflammation.
  confidence: medium
  date: 2026-10-14
  sources: [S1, S3]
  license: CC BY 4.0
  note: NSAID class is on the label; the inflammation statement is from MedlinePlus only.
- id: F3
  claim: NSAIDs like ibuprofen can cause severe stomach bleeding; risk is higher with use beyond directions, older age, heavy daily alcohol, prior ulcers/bleeding, blood thinners or steroids, or other NSAIDs.
  confidence: high
  date: 2026-10-14
  sources: [S1, S3]
  license: CC BY 4.0
  note: Label gives age 60+ and 3+ drinks/day; MedlinePlus says "older in age" and "large amounts of alcohol" and adds smoking and poor health.
- id: F4
  claim: NSAIDs other than aspirin raise the risk of heart attack and stroke (label also lists heart failure), more so at higher doses or longer use.
  confidence: high
  date: 2026-10-14
  sources: [S1, S3]
  license: CC BY 4.0
- id: F5
  claim: Avoid ibuprofen at 20 weeks of pregnancy or later unless a doctor specifically says to.
  confidence: high
  date: 2026-10-14
  sources: [S1, S3]
  license: CC BY 4.0
- id: F6
  claim: Acetaminophen relieves mild-to-moderate pain and reduces fever; it is not an NSAID (MedlinePlus classes it as an analgesic/antipyretic).
  confidence: high
  date: 2026-10-14
  sources: [S2, S4]
  license: CC BY 4.0
- id: F7
  claim: Severe liver damage can occur with more than 4,000 mg of acetaminophen in 24 hours; risk rises when combined with other acetaminophen products or 3+ alcoholic drinks a day.
  confidence: high
  date: 2026-10-14
  sources: [S2, S4]
  license: CC BY 4.0
  note: Alcohol threshold is from the label.
- id: F8
  claim: Do not take two products containing acetaminophen at once; check labels for it.
  confidence: high
  date: 2026-10-14
  sources: [S2, S4]
  license: CC BY 4.0
- id: F9
  claim: For both, stop and ask a doctor if pain lasts over 10 days or fever over 3 days, or if they get worse.
  confidence: high
  date: 2026-10-14
  sources: [S2, S3, S4]
  license: CC BY 4.0
- id: F10
  claim: Ibuprofen users should stop and seek help for signs of stomach bleeding (fainting, bloody or black stools, vomiting blood) or heart/stroke symptoms (chest pain, slurred speech, trouble breathing, one-sided weakness).
  confidence: high
  date: 2026-10-14
  sources: [S1, S3]
  license: CC BY 4.0

## Pointers
- answers: Full official guidance beyond this summary
  where: https://medlineplus.gov/druginfo/meds/a682159.html; https://medlineplus.gov/druginfo/meds/a681004.html; https://dailymed.nlm.nih.gov/dailymed/lookup.cfm?setid=00653b7c-7099-487e-9a01-e89781c21323; https://dailymed.nlm.nih.gov/dailymed/lookup.cfm?setid=0005aafa-08f8-4990-a54d-cc097195708f
  access: open, no account

## Open issues (pre-review)
- Labels are one manufacturer each (generic Drug Facts); wording across brands is FDA-standardized but not checked for every product.

---

## Identity theft: signs, protection, and reporting

- URL: https://iinventedlight.com/library/reference/us-gov/identity-theft/
- Markdown: https://iinventedlight.com/library/reference/us-gov/identity-theft.md
- License: CC-BY-4.0 (source: https://consumer.ftc.gov/articles/what-know-about-identity-theft)
- checked_on: 2026-10-14
- status: reviewed

# Identity theft: signs, protection, and reporting

## Summary
Identity theft is someone using your personal or financial information without permission. Watch bills, bank statements, and credit reports. A credit freeze and fraud alert are free. Report at IdentityTheft.gov for a personal recovery plan.

> **General information, not legal advice. Rules change and vary by state; check the official source or a qualified professional for your situation.**

## Facts
- id: F1
  claim: Identity theft means someone uses your personal or financial information without permission.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F2
  claim: Signs include missing bills, charges you did not make, unexpected withdrawals, and unfamiliar accounts on your credit report.
  confidence: high
  date: 2026-10-14
  sources: [S1, S2]
  license: CC BY 4.0
- id: F3
  claim: A credit freeze blocks new credit accounts and is free to place and lift; contact each of Equifax, Experian, and TransUnion.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F4
  claim: A free one-year fraud alert can be placed by contacting one bureau, which must notify the other two.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F5
  claim: The IRS, your bank, and your employer will not call, email, or text asking for your Social Security number.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F6
  claim: Report identity theft at IdentityTheft.gov (Spanish: RobodeIdentidad.gov) to get a free personal recovery plan.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0

## Pointers
- answers: Full official guidance beyond this summary
  where: https://consumer.ftc.gov/articles/what-know-about-identity-theft; https://consumer.ftc.gov/articles/free-credit-reports
  access: open, no account

---

## IRA contribution limits

- URL: https://iinventedlight.com/library/reference/us-gov/ira-contribution-limits/
- Markdown: https://iinventedlight.com/library/reference/us-gov/ira-contribution-limits.md
- License: CC-BY-4.0 (source: https://www.irs.gov/retirement-plans/plan-participant-employee/retirement-topics-ira-contribution-limits)
- checked_on: 2026-10-14
- status: reviewed

# IRA contribution limits

## Summary
For 2026, total contributions to all traditional and Roth IRAs are capped at $7,500 ($8,600 at age 50+), or your taxable compensation if less. 2024–2025: $7,000 ($8,000 at 50+). Roth contributions may be further limited by income.

> **General information, not financial or tax advice. Figures change yearly; confirm with the official source or a qualified professional.**

## Facts
- id: F1
  claim: 2026 combined limit for all traditional and Roth IRAs: $7,500, or $8,600 at age 50 or older, or taxable compensation if lower.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F2
  claim: 2024 and 2025 limit: $7,000, or $8,000 at age 50 or older.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F3
  claim: Roth IRA contributions may be reduced based on filing status and income.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F4
  claim: Since 2020 there is no age limit on regular contributions to traditional or Roth IRAs.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F5
  claim: A spouse filing jointly may contribute even without their own taxable compensation (spousal IRA), within the couple's combined compensation.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F6
  claim: Excess contributions are taxed 6% per year while they stay in the IRA.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0

## Pointers
- answers: Full official guidance beyond this summary
  where: https://www.irs.gov/retirement-plans/plan-participant-employee/retirement-topics-ira-contribution-limits
  access: open, no account

---

## Becoming a U.S. citizen (naturalization)

- URL: https://iinventedlight.com/library/reference/us-gov/naturalization-basics/
- Markdown: https://iinventedlight.com/library/reference/us-gov/naturalization-basics.md
- License: CC-BY-4.0 (source: https://www.uscis.gov/citizenship/learn-about-citizenship/citizenship-and-naturalization/i-am-a-lawful-permanent-resident-of-5-years)
- checked_on: 2026-10-14
- status: reviewed

# Becoming a U.S. citizen (naturalization)

## Summary
Most people naturalize after 5 years as a green-card holder by filing Form N-400. Core requirements: age 18+, continuous residence, 30 months physical presence, good moral character, English and civics tests, and the Oath of Allegiance.

> **General information, not legal advice. Rules change; confirm with the agency or an attorney.**

## Facts
- id: F1
  claim: Naturalization is the voluntary process by which a lawful permanent resident becomes a U.S. citizen.
  confidence: high
  date: 2026-10-14
  sources: [S2, S3]
  license: CC BY 4.0
  checked_on: 2026-10-14
- id: F2
  claim: The most common path is 5 years as a lawful permanent resident; other paths exist for spouses of citizens, military members and children of citizens.
  confidence: medium
  date: 2026-10-14
  sources: [S1, S2]
  license: CC BY 4.0
  checked_on: 2026-10-14
- id: F3
  claim: The application is Form N-400; applicants must be at least 18 when filing.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
  checked_on: 2026-10-14
- id: F4
  claim: Applicants need 5 years' continuous residence and at least 30 months physically present in the U.S. in the 5 years before filing, plus 3 months in the filing state/district.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
  checked_on: 2026-10-14
- id: F5
  claim: Applicants must show good moral character, basic English, civics knowledge, attachment to the Constitution, and take the Oath of Allegiance.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
  checked_on: 2026-10-14
- id: F6
  claim: Some older long-term residents are exempt from the English test and may take civics in their language.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
  checked_on: 2026-10-14

## Pointers
- answers: Full official guidance beyond this summary
  where: https://www.uscis.gov/citizenship/learn-about-citizenship/citizenship-and-naturalization/i-am-a-lawful-permanent-resident-of-5-years; https://www.uscis.gov/citizenship/learn-about-citizenship/citizenship-and-naturalization; https://www.usa.gov/become-us-citizen
  access: open, no account

---

## Overtime pay under federal law

- URL: https://iinventedlight.com/library/reference/us-gov/overtime-pay/
- Markdown: https://iinventedlight.com/library/reference/us-gov/overtime-pay.md
- License: CC-BY-4.0 (source: https://www.dol.gov/agencies/whd/overtime)
- checked_on: 2026-10-14
- status: reviewed

# Overtime pay under federal law

## Summary
Under the FLSA, covered non-exempt employees get at least 1.5× their regular rate for hours over 40 in a workweek. Weekends and holidays do not require overtime by themselves, and hours cannot be averaged across weeks.

> **General information, not legal advice. State laws may give more protection; check your state labor department.**

## Facts
- id: F1
  claim: Covered, non-exempt employees must get at least time and a half their regular rate for hours over 40 in a workweek.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F2
  claim: The FLSA sets no limit on weekly hours for employees 16 and older.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F3
  claim: The FLSA does not require overtime pay for weekends, holidays, or rest days unless hours exceed 40 that week.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F4
  claim: A workweek is a fixed, recurring 168-hour period that can start on any day and hour.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F5
  claim: Hours cannot be averaged over two or more weeks.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0

## Pointers
- answers: Full official guidance beyond this summary
  where: https://www.dol.gov/agencies/whd/overtime
  access: open, no account

---

## U.S. passport processing times

- URL: https://iinventedlight.com/library/reference/us-gov/passport-processing-times/
- Markdown: https://iinventedlight.com/library/reference/us-gov/passport-processing-times.md
- License: CC-BY-4.0 (source: https://travel.state.gov/content/travel/en/passports/how-apply/processing-times.html)
- checked_on: 2026-10-14
- status: reviewed

# U.S. passport processing times

## Summary
As of 2026-10-14, routine passport processing takes 4–6 weeks and expedited takes 2–3 weeks (extra $60), not counting up to 2 weeks of mailing each way. Urgent travel within 14 days needs a passport agency appointment.

> **General information, not legal or travel advice. Rules, fees, and processing times change; confirm with travel.state.gov, TSA, or CBP before you travel.**

## Facts
- id: F1
  claim: Routine processing takes 4–6 weeks.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F2
  claim: Expedited processing takes 2–3 weeks for an extra $60.
  confidence: medium
  date: 2026-10-14
  sources: [S1, S2]
  license: CC BY 4.0
- id: F3
  claim: Mailing is not included; allow up to 2 weeks for the application to arrive and up to 2 weeks for the passport to reach you.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F4
  claim: Urgent travel requires a passport agency appointment and proof of international travel within 14 calendar days.
  confidence: medium
  date: 2026-10-14
  sources: [S1, S2]
  license: CC BY 4.0
- id: F5
  claim: The busiest period is late winter through summer; State recommends applying October–December.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F6
  claim: Status updates are emailed and available at passportstatus.state.gov.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0

## Pointers
- answers: Current times (change often)
  where: https://travel.state.gov/content/travel/en/passports/how-apply/processing-times.html
  access: open

---

## Renewing a U.S. passport (mail or online)

- URL: https://iinventedlight.com/library/reference/us-gov/passport-renewal/
- Markdown: https://iinventedlight.com/library/reference/us-gov/passport-renewal.md
- License: CC-BY-4.0 (source: https://travel.state.gov/content/travel/en/passports/have-passport/renew.html)
- checked_on: 2026-10-14
- status: reviewed

# Renewing a U.S. passport (mail or online)

## Summary
Adults can renew by mail with Form DS-82 if their most recent passport is undamaged, was issued within the last 15 years when they were 16 or older, and is in their current name (or they have a name-change document). The fee is $130 for a book, $30 for a card, or $160 for both; expedited service costs $60 more.

> **General information, not legal or travel advice. Rules, fees, and processing times change; confirm with travel.state.gov, TSA, or CBP before you travel.**

## Facts
- id: F1
  claim: You can renew by mail if your most recent passport was issued within 15 years, when you were 16 or older, was valid for 10 years, is undamaged, and was never reported lost or stolen.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F2
  claim: Children under 16 cannot renew; they must apply in person for a new passport.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F3
  claim: Mail renewals use Form DS-82, printed single-sided, signed and dated.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F4
  claim: Renewal fees are $130 (book), $30 (card), or $160 (book and card), paid by check or money order to the U.S. Department of State.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F5
  claim: Expedited service adds $60; 1–3 day return delivery adds $26.87.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F6
  claim: You must send your most recent passport and one photo; it is returned separately.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F7
  claim: If your name changed, include a certified copy of the marriage certificate, divorce decree, or court order.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F8
  claim: Eligible U.S. citizens wanting routine service can renew online; fees are the same ($130 book, $30 card, $160 both).
  confidence: medium
  date: 2026-10-14
  sources: [S1, S2]
  license: CC BY 4.0
- id: F9
  claim: Online renewal requires being 25 or older, a 10-year passport expiring within 1 year or expired less than 5 years ago, no name or sex change, being in a U.S. state or territory, and no travel for at least 6 weeks.
  confidence: medium
  date: 2026-10-14
  sources: [S2]
  license: CC BY 4.0
- id: F10
  claim: Online renewals cannot be expedited, and the old passport is cancelled once you apply.
  confidence: medium
  date: 2026-10-14
  sources: [S2]
  license: CC BY 4.0
- id: F11
  claim: The only official online renewal site is opr.travel.state.gov; third-party sites may be fraudulent.
  confidence: medium
  date: 2026-10-14
  sources: [S2]
  license: CC BY 4.0

## Pointers
- answers: Official online renewal (only authorized site)
  where: https://opr.travel.state.gov
  access: open; applicant must submit their own application

---

## Federal Pell Grant: 2026–27 amounts

- URL: https://iinventedlight.com/library/reference/us-gov/pell-grant/
- Markdown: https://iinventedlight.com/library/reference/us-gov/pell-grant.md
- License: CC-BY-4.0 (source: https://fsapartners.ed.gov/knowledge-center/library/dear-colleague-letters/2026-01-30/2026-27-federal-pell-grant-maximum-and-minimum-award-amounts)
- checked_on: 2026-10-14
- status: reviewed

# Federal Pell Grant: 2026–27 amounts

## Summary
For the 2026–27 award year (July 1, 2026–June 30, 2027), the maximum Federal Pell Grant is $7,395 and the minimum is $740. Pell Grants generally do not have to be repaid; you apply by submitting the FAFSA.

> **General information, not financial, legal, or academic advice. Aid rules and amounts change each award year; confirm at StudentAid.gov or with your school's financial aid office.**

## Facts
- id: F1
  claim: The 2026–27 maximum Pell Grant award is $7,395.
  confidence: medium
  date: 2026-10-14
  sources: [S1, S2]
  license: CC BY 4.0
- id: F2
  claim: The 2026–27 minimum Pell Grant award is $740 (10% of the maximum, rounded to the nearest $5).
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F3
  claim: The 2026–27 award year runs July 1, 2026 through June 30, 2027.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F4
  claim: Starting in 2026–27, a student whose Student Aid Index (SAI) is at least twice the maximum Pell award ($14,790 for 2026–27) cannot get a Pell Grant, with an exception for certain dependents of deceased service members and public safety officers.
  confidence: medium
  date: 2026-10-14
  sources: [S1, S3]
  license: CC BY 4.0
- id: F5
  claim: A student may receive up to 150% of their scheduled Pell award in one award year in some cases.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F6
  claim: You must complete the FAFSA to receive federal aid, including a Pell Grant; the Federal Student Aid Estimator only gives an estimate.
  confidence: medium
  date: 2026-10-14
  sources: [S2]
  license: CC BY 4.0
- id: F7
  claim: Pell Grants are funds you do not have to pay back.
  confidence: medium
  date: 2026-10-14
  sources: [S2]
  license: CC BY 4.0
- id: F8
  claim: The 2027–28 maximum is expected to be announced in early 2027.
  confidence: medium
  date: 2026-10-14
  sources: [S2]
  license: CC BY 4.0

## Pointers
- answers: Pell eligibility formulas and tables
  where: 2026–27 SAI and Pell Grant Eligibility Guide, fsapartners.ed.gov
  access: open

---

## Physical activity for adults

- URL: https://iinventedlight.com/library/reference/us-gov/physical-activity-adults/
- Markdown: https://iinventedlight.com/library/reference/us-gov/physical-activity-adults.md
- License: CC-BY-4.0 (source: https://www.cdc.gov/physical-activity-basics/guidelines/adults.html)
- checked_on: 2026-10-14
- status: reviewed

# Physical activity for adults

## Summary
Adults need at least 150 minutes a week of moderate aerobic activity (or 75 minutes vigorous, or a mix) plus muscle-strengthening on 2+ days. It can be split into short chunks; any activity beats none.

> **General information, not medical advice. See a clinician about your own situation; in an emergency call 911 (US).**

## Facts
- id: F1
  claim: Adults need at least 150 minutes of moderate-intensity aerobic activity per week, e.g. 30 minutes on 5 days.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
  checked_on: 2026-10-14
- id: F2
  claim: 75 minutes a week of vigorous activity, or an equivalent moderate/vigorous mix, also meets the aerobic target.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
  checked_on: 2026-10-14
- id: F3
  claim: Adults should also do muscle-strengthening work on at least 2 days a week covering all major muscle groups.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
  checked_on: 2026-10-14
- id: F4
  claim: Activity can be broken into shorter sessions spread across the week.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
  checked_on: 2026-10-14
- id: F5
  claim: Some activity is better than none; going beyond 150 minutes brings additional benefit.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
  checked_on: 2026-10-14

## Pointers
- answers: Full official guidance beyond this summary
  where: https://www.cdc.gov/physical-activity-basics/guidelines/adults.html
  access: open, no account

---

## REAL ID basics

- URL: https://iinventedlight.com/library/reference/us-gov/real-id/
- Markdown: https://iinventedlight.com/library/reference/us-gov/real-id.md
- License: CC-BY-4.0 (source: https://www.dhs.gov/real-id/real-id-faqs)
- checked_on: 2026-10-14
- status: reviewed

# REAL ID basics

## Summary
Since May 7, 2025, a state license or ID must be REAL ID-compliant (usually a star marking) to board domestic flights or enter certain federal facilities. A passport or other accepted ID works instead. Children under 18 do not need ID from TSA for domestic travel.

> **General information, not legal advice. Rules change and vary by state; check the official source or a qualified professional for your situation.**

## Facts
- id: F1
  claim: REAL ID enforcement began on May 7, 2025.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F2
  claim: A REAL ID-compliant license/ID or another acceptable document is needed to board commercial aircraft, access federal facilities, and enter nuclear power plants.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F3
  claim: Compliant cards carry a marking (usually a star) on the upper part of the card.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F4
  claim: Enhanced Driver's Licenses from Washington, Michigan, Minnesota, New York, and Vermont are accepted alternatives, even without a star.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F5
  claim: For domestic flights you need one valid ID, either a REAL ID or an alternative such as a passport, not both; international travel still needs a passport.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F6
  claim: TSA does not require children under 18 to show ID for travel within the US.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0

## Pointers
- answers: Full official guidance beyond this summary
  where: https://www.dhs.gov/real-id/real-id-faqs
  access: open, no account

---

## How to recognize and report phishing

- URL: https://iinventedlight.com/library/reference/us-gov/recognize-phishing/
- Markdown: https://iinventedlight.com/library/reference/us-gov/recognize-phishing.md
- License: CC-BY-4.0 (source: https://consumer.ftc.gov/articles/how-recognize-and-avoid-phishing-scams)
- checked_on: 2026-10-14
- status: reviewed

# How to recognize and report phishing

## Summary
Phishing messages pose as a trusted company and push you to click a link, open an attachment, or "confirm" information. Contact the company through a number or site you already know. Forward phishing email to reportphishing@apwg.org, texts to 7726, and report to ReportFraud.ftc.gov.

> **General information. Follow the official source for full instructions.**

## Facts
- id: F1
  claim: Common phishing stories: suspicious activity on your account, a payment problem, a need to confirm personal information, an unrecognized invoice, a government refund, or a free coupon.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F2
  claim: Warning signs include a generic greeting and a request to click a link to update payment details.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F3
  claim: If the message seems to come from a company you deal with, contact it using a phone number or website you know is real, not the message's links.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F4
  claim: Protections: auto-updating security software and phone software, multi-factor authentication, and backups.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F5
  claim: Report phishing email to reportphishing@apwg.org, phishing texts to 7726 (SPAM), and either to ReportFraud.ftc.gov.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F6
  claim: If you gave a scammer your information, IdentityTheft.gov lists steps based on what was exposed.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0

## Pointers
- answers: Full official guidance beyond this summary
  where: https://consumer.ftc.gov/articles/how-recognize-and-avoid-phishing-scams; https://www.cisa.gov/secure-our-world/use-strong-passwords
  access: open, no account

---

## How to register to vote (US)

- URL: https://iinventedlight.com/library/reference/us-gov/register-to-vote/
- Markdown: https://iinventedlight.com/library/reference/us-gov/register-to-vote.md
- License: CC-BY-4.0 (source: https://www.usa.gov/register-to-vote)
- checked_on: 2026-10-14
- status: reviewed

# How to register to vote (US)

## Summary
Start at vote.gov and pick your state; depending on state rules you can register online, by mail, or in person. The National Mail Voter Registration Form is another option. Overseas and military voters use the Federal Voting Assistance Program.

> **General information, not legal advice. Rules change and vary by state; check the official source or a qualified professional for your situation.**

## Facts
- id: F1
  claim: vote.gov sends you to your state's instructions for registering online, by mail, or in person.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F2
  claim: The National Mail Voter Registration Form can be filled in, printed, signed, and mailed to the address listed for your state.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F3
  claim: You may also be able to register at the DMV or at public assistance offices such as SNAP or TANF offices.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F4
  claim: Citizens abroad and service members and their families register and request absentee ballots through the Federal Voting Assistance Program.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F5
  claim: Required documents vary by state; usually a driver's license or state ID, sometimes alternatives like a bank statement or utility bill.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0

## Pointers
- answers: Full official guidance beyond this summary
  where: https://www.usa.gov/register-to-vote
  access: open, no account

---

## How to renew a US passport by mail

- URL: https://iinventedlight.com/library/reference/us-gov/renew-us-passport/
- Markdown: https://iinventedlight.com/library/reference/us-gov/renew-us-passport.md
- License: CC-BY-4.0 (source: https://travel.state.gov/content/travel/en/passports/have-passport/renew.html)
- checked_on: 2026-10-14
- status: reviewed

# How to renew a US passport by mail

## Summary
Adults can renew by mail with Form DS-82 if their last passport was issued within 15 years, at age 16 or older, was valid for 10 years, is undamaged, and was never reported lost or stolen. Book fee $130, card $30, both $160; expedited service adds $60. Eligible routine applicants may renew online.

> **General information. Follow the official source for full instructions.**

## Facts
- id: F1
  claim: You can renew by mail if you can send your most recent passport, it was issued in the last 15 years, at age 16 or older, valid for 10 years, never reported lost or stolen, undamaged, and in your current name (or you have a name-change document).
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F2
  claim: If you do not meet those requirements, including lost/stolen passports or ones issued over 15 years ago, you must apply in person.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F3
  claim: Passports for children under 16 cannot be renewed; a new application is required.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F4
  claim: Mail renewals use Form DS-82, printed single-sided, with one photo stapled on.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F5
  claim: DS-82 application fees: passport book $130, passport card $30, both $160, paid by check or money order to the U.S. Department of State.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
  note: Fees as checked 2026-10-14; may change.
- id: F6
  claim: Expedited service (2–3 weeks) costs an extra $60; 1–3 day return shipping costs an extra $26.87.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F7
  claim: Eligible applicants seeking routine service can renew online instead.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0

## Pointers
- answers: Full official guidance beyond this summary
  where: https://travel.state.gov/content/travel/en/passports/have-passport/renew.html
  access: open, no account

---

## Replacing a Social Security card

- URL: https://iinventedlight.com/library/reference/us-gov/replace-social-security-card/
- Markdown: https://iinventedlight.com/library/reference/us-gov/replace-social-security-card.md
- License: CC-BY-4.0 (source: https://www.ssa.gov/number-card/replace-card)
- checked_on: 2026-10-14
- status: reviewed

# Replacing a Social Security card

## Summary
A replacement card is free. Start online (fastest); otherwise make an appointment at a local office. The card arrives by mail in 5–10 business days. You often do not need the physical card if you know your number.

> **General information, not legal advice. Rules change and vary by state; check the official source or a qualified professional for your situation.**

## Facts
- id: F1
  claim: A replacement Social Security card is free.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F2
  claim: Some people can apply online; others need an appointment at a local Social Security office.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F3
  claim: Starting online is the fastest way; local offices cannot print a card on the spot.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F4
  claim: The card arrives by mail 5 to 10 business days after the request is completed.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F5
  claim: In most cases a physical card is not needed if you know your number.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F6
  claim: SSA phone line: 1-800-772-1213 (TTY 1-800-325-0778), weekdays 8 a.m.–7 p.m.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0

## Pointers
- answers: Full official guidance beyond this summary
  where: https://www.ssa.gov/number-card/replace-card
  access: open, no account

---

## Reporting an unsafe product (SaferProducts.gov)

- URL: https://iinventedlight.com/library/reference/us-gov/report-unsafe-product/
- Markdown: https://iinventedlight.com/library/reference/us-gov/report-unsafe-product.md
- License: CC-BY-4.0 (source: https://www.saferproducts.gov/)
- checked_on: 2026-10-14
- status: reviewed

# Reporting an unsafe product (SaferProducts.gov)

## Summary
Report a dangerous consumer product to CPSC at SaferProducts.gov. CPSC staff review every report, and reports go into a searchable public database that you can also check before buying.

> **General information, not legal advice. Recall and warranty rights depend on the product and your state; confirm with the official source.**

## Facts
- id: F1
  claim: SaferProducts.gov is CPSC's site for reporting harm or potential harm from consumer products.
  confidence: medium
  date: 2026-10-14
  sources: [S1, S2]
  license: CC BY 4.0
- id: F2
  claim: CPSC investigators and product safety experts review each report.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F3
  claim: A report may contribute to CPSC seeking a recall, penalties, or new safety rules.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F4
  claim: Reports are published in a searchable public database, and the data can be downloaded.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F5
  claim: Manufacturers, importers and labelers are notified about reports on their products and can respond; their comments may be published.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F6
  claim: Spanish speakers can report by email to clearinghouse@cpsc.gov or call 800-638-2772 (press 2).
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0

## Pointers
- answers: File a report
  where: https://www.saferproducts.gov/
  access: open form

---

## Safe minimum cooking temperatures

- URL: https://iinventedlight.com/library/reference/us-gov/safe-cooking-temperatures/
- Markdown: https://iinventedlight.com/library/reference/us-gov/safe-cooking-temperatures.md
- License: CC-BY-4.0 (source: https://www.foodsafety.gov/food-safety-charts/safe-minimum-internal-temperatures)
- checked_on: 2026-10-14
- status: reviewed

# Safe minimum cooking temperatures

## Summary
Use a food thermometer. Poultry and leftovers: 165°F (74°C). Ground meat: 160°F (71°C). Whole cuts of beef, pork, lamb, veal: 145°F (63°C) plus a 3-minute rest. Fish: 145°F (63°C).

> **General information. Follow the official source for full instructions.**

## Facts
- id: F1
  claim: Checking with a food thermometer is the recommended way to know meat has reached a safe internal temperature.
  confidence: high
  date: 2026-10-14
  sources: [S1, S2]
  license: CC BY 4.0
- id: F2
  claim: All poultry (whole, parts, ground, and stuffing inside the bird) should reach 165°F (74°C).
  confidence: high
  date: 2026-10-14
  sources: [S1, S2]
  license: CC BY 4.0
- id: F3
  claim: Ground beef, pork, lamb, and veal, and sausage, should reach 160°F (71°C).
  confidence: high
  date: 2026-10-14
  sources: [S1, S2]
  license: CC BY 4.0
- id: F4
  claim: Steaks, roasts, and chops of beef, pork, lamb, and veal should reach 145°F (63°C), then rest at least 3 minutes.
  confidence: high
  date: 2026-10-14
  sources: [S1, S2]
  license: CC BY 4.0
- id: F5
  claim: Fish should reach 145°F (63°C), or cook until the flesh is opaque and flakes easily with a fork.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F6
  claim: Egg dishes such as quiche should reach 160°F (71°C); raw eggs should be cooked until yolk and white are firm.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F7
  claim: Leftovers and casseroles should be reheated to 165°F (74°C).
  confidence: high
  date: 2026-10-14
  sources: [S1, S2]
  license: CC BY 4.0

## Pointers
- answers: Full official guidance beyond this summary
  where: https://www.foodsafety.gov/food-safety-charts/safe-minimum-internal-temperatures; https://www.fsis.usda.gov/food-safety/safe-food-handling-and-preparation/food-safety-basics/leftovers-and-food-safety
  access: open, no account

---

## US travelers in Europe: Schengen, EES and ETIAS

- URL: https://iinventedlight.com/library/reference/us-gov/schengen-etias/
- Markdown: https://iinventedlight.com/library/reference/us-gov/schengen-etias.md
- License: CC-BY-4.0 (source: https://travel.state.gov/en/international-travel/planning/guidance/europe.html)
- checked_on: 2026-10-14
- status: reviewed

# US travelers in Europe: Schengen, EES and ETIAS

## Summary
With a US passport you can visit the Schengen area visa-free for up to 90 days in any 180-day period; your passport must be valid 3 months past departure. EES records fingerprints/face at the border. ETIAS (online pre-travel authorization, EUR 20) is not yet running and has no official start date; the EU says it will announce the date several months before launch.

> **General information. Entry rules change; confirm with travel.state.gov and official EU sites before you travel.**

## Facts
- id: F1
  claim: US passport holders may stay in the Schengen area up to 90 days in any 180-day period for tourism or business.
  confidence: high
  date: 2026-10-14
  sources: [S1, S2]
  license: CC BY 4.0
  checked_on: 2026-10-14
- id: F2
  claim: Your passport must be valid at least 3 months beyond your planned departure from the Schengen area.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
  checked_on: 2026-10-14
- id: F3
  claim: Stays over 90 days need a national visa from the country where you will spend most time.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
  checked_on: 2026-10-14
- id: F4
  claim: The EU Entry/Exit System (EES) collects fingerprints, a facial image and entry/exit dates at external borders; the EU says it has been fully operational since 2026-04-10.
  confidence: high
  date: 2026-10-14
  sources: [S1, S3]
  license: CC BY 4.0
  checked_on: 2026-10-14
- id: F5
  claim: ETIAS is not operating as of 2026-10-14 and no applications are collected; the EU will announce the start date several months before launch. An April 2026 EU article had targeted the last quarter of 2026, but the official ETIAS pages no longer give a date.
  confidence: high
  date: 2026-10-14
  sources: [S2, S3, S4]
  license: CC BY 4.0
  checked_on: 2026-10-14
- id: F6
  claim: Once live, an ETIAS authorization is applied for online (official website or app only), costs a one-time EUR 20 fee, covers 30 European countries, and is valid 3 years or until the passport expires, whichever comes first.
  confidence: high
  date: 2026-10-14
  sources: [S3, S4]
  license: CC BY 4.0
  checked_on: 2026-10-14

## Pointers
- answers: Full official guidance beyond this summary
  where: https://travel.state.gov/en/international-travel/planning/guidance/europe.html; https://travel-europe.europa.eu/etias/about-etias/what-is-etias.html; https://home-affairs.ec.europa.eu/policies/schengen/smart-borders/european-travel-information-authorisation-system_en; https://home-affairs.ec.europa.eu/news/main-differences-between-ees-and-etias-what-travellers-need-know-2026-04-28_en
  access: open, no account

## Disagreements
EES 'as of October 12, 2025' (State, start of phased rollout) vs. 'fully operational since 10 April 2026' (EU): consistent—phased start, then full operation.

ETIAS launch: S3 (2026-04-28) said 'scheduled to start in the last quarter of 2026'; S2 and S4 (checked 2026-10-14) say ETIAS is not in operation and the date will be announced several months before launch. The later official wording is used. EES 'no fee' was dropped: not stated on the EU pages checked.

---

## Sleep

- URL: https://iinventedlight.com/library/reference/us-gov/sleep/
- Markdown: https://iinventedlight.com/library/reference/us-gov/sleep.md
- License: CC-BY-4.0 (source: https://www.cdc.gov/sleep/about/index.html)
- checked_on: 2026-10-14
- status: reviewed

# Sleep

## Summary
Most adults need 7 or more hours of sleep a night (CDC: 7–9 for ages 61–64, 7–8 for 65+), and needs are higher for children and teens. Regular schedules and a quiet, cool, screen-free bedroom help; ongoing sleep problems are worth raising with a clinician.

> **General information, not medical advice. See a clinician about your own situation; in an emergency call 911 (US).**

## Facts
- id: F1
  claim: CDC recommended sleep: adults 18–60 7+ hours; 61–64 7–9; 65+ 7–8; teens 13–17 8–10; ages 6–12 9–12.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F2
  claim: Enough sleep is linked to fewer illnesses, better mood and attention, and lower risk of type 2 diabetes, heart disease, high blood pressure, and stroke.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F3
  claim: Helpful habits: consistent bed and wake times, a quiet cool bedroom, devices off 30+ minutes before bed, avoiding big meals, alcohol, and late caffeine, and regular exercise.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F4
  claim: Common sleep disorders include insomnia, restless legs syndrome, narcolepsy, and sleep apnea; a provider can test for them, sometimes with a sleep study.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0

## Pointers
- answers: Full official guidance and details beyond this summary
  where: https://www.cdc.gov/sleep/about/index.html
  access: open, no account

## Open issues (pre-review)
- Single open source (CDC). Add a second (e.g. NHLBI) before review.

---

## Social Security full retirement age and early benefits

- URL: https://iinventedlight.com/library/reference/us-gov/social-security-retirement-age/
- Markdown: https://iinventedlight.com/library/reference/us-gov/social-security-retirement-age.md
- License: CC-BY-4.0 (source: https://www.ssa.gov/benefits/retirement/planner/agereduction.html)
- checked_on: 2026-10-14
- status: reviewed

# Social Security full retirement age and early benefits

## Summary
Benefits can start at 62 but are reduced; full retirement age is 67 for people born in 1960 or later (66 to 66 and 10 months for 1943–1959). Delaying past full retirement age up to 70 increases the benefit.

> **General information, not financial or tax advice. Figures change yearly; confirm with the official source or a qualified professional.**

## Facts
- id: F1
  claim: Retirement benefits can start as early as age 62, but full benefits begin only at full retirement age.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F2
  claim: Full retirement age is 66 for birth years 1943–1954, rises by 2 months per year for 1955–1959, and is 67 for 1960 and later.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F3
  claim: For people with a full retirement age of 67, starting at 62 cuts the benefit by about 30%.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F4
  claim: Delaying benefits past full retirement age, up to 70, earns delayed retirement credits that raise the monthly amount.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F5
  claim: People born on January 1 use the previous year's row in the full-retirement-age table.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F6
  claim: Even if delaying benefits, apply for Medicare within 3 months of turning 65 to avoid possibly higher Part B and Part D costs.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0

## Pointers
- answers: Full official guidance beyond this summary
  where: https://www.ssa.gov/benefits/retirement/planner/agereduction.html
  access: open, no account

---

## Planets of the solar system

- URL: https://iinventedlight.com/library/reference/us-gov/solar-system-planets/
- Markdown: https://iinventedlight.com/library/reference/us-gov/solar-system-planets.md
- License: CC-BY-4.0 (source: https://science.nasa.gov/solar-system/planets/)
- checked_on: 2026-10-14
- status: reviewed

# Planets of the solar system

## Summary
The solar system has eight planets: Mercury, Venus, Earth, Mars, Jupiter, Saturn, Uranus and Neptune. Pluto is one of five officially recognized dwarf planets (Ceres, Pluto, Haumea, Makemake, Eris). Jupiter is largest; Mercury is smallest and closest to the Sun.

> **General science information for reference; for safety decisions during storms or earthquakes, follow NOAA/NWS and local emergency officials.**

## Facts
- id: F1
  claim: "The solar system has eight planets: Mercury, Venus, Earth, Mars, Jupiter, Saturn, Uranus, and Neptune."
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F2
  claim: "There are five officially recognized dwarf planets: Ceres, Pluto, Haumea, Makemake, and Eris."
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F3
  claim: Pluto's reclassification as a dwarf planet reduced the planet count from nine to eight.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F4
  claim: Mercury, Venus, Earth and Mars are terrestrial planets with solid surfaces.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F5
  claim: Jupiter and Saturn are gas giants; Uranus and Neptune are ice giants.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F6
  claim: Jupiter is the largest planet; Mercury is the smallest and closest to the Sun.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F7
  claim: Earth is the third planet from the Sun and the fifth largest.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0

## Pointers
- answers: Per-planet facts
  where: https://science.nasa.gov/solar-system/planets/
  access: open

---

## Standard deduction, tax year 2026

- URL: https://iinventedlight.com/library/reference/us-gov/standard-deduction-2026/
- Markdown: https://iinventedlight.com/library/reference/us-gov/standard-deduction-2026.md
- License: CC-BY-4.0 (source: https://www.irs.gov/newsroom/irs-releases-tax-inflation-adjustments-for-tax-year-2026-including-amendments-from-the-one-big-beautiful-bill)
- checked_on: 2026-10-14
- status: reviewed

# Standard deduction, tax year 2026

## Summary
For tax year 2026 (returns filed in 2027): $16,100 single or married filing separately, $24,150 head of household, $32,200 married filing jointly or surviving spouse.

> **General information, not financial or tax advice. Check IRS/agency sources or a professional for your situation.**

## Facts
- id: F1
  claim: 2026 standard deduction: $16,100 for single filers and married filing separately.
  confidence: high
  date: 2026-10-14
  sources: [S1, S2]
  license: CC BY 4.0
  checked_on: 2026-10-14
- id: F2
  claim: 2026 standard deduction: $24,150 for heads of household.
  confidence: high
  date: 2026-10-14
  sources: [S1, S2]
  license: CC BY 4.0
  checked_on: 2026-10-14
- id: F3
  claim: 2026 standard deduction: $32,200 for married filing jointly and surviving spouses.
  confidence: high
  date: 2026-10-14
  sources: [S1, S2]
  license: CC BY 4.0
  checked_on: 2026-10-14
- id: F4
  claim: 2025 amounts (filed in 2026) were $15,750 / $23,625 / $31,500 respectively.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
  checked_on: 2026-10-14
- id: F5
  claim: Taxpayers 65+ or blind get an additional standard deduction: $1,650 each for joint filers, $2,050 for single/head of household in 2026.
  confidence: medium
  date: 2026-10-14
  sources: [S2]
  license: CC BY 4.0
  checked_on: 2026-10-14
- id: F6
  claim: You subtract either the standard deduction or itemized deductions, whichever is larger.
  confidence: medium
  date: 2026-10-14
  sources: [S2]
  license: CC BY 4.0
  checked_on: 2026-10-14

## Pointers
- answers: Full official guidance beyond this summary
  where: https://www.irs.gov/newsroom/irs-releases-tax-inflation-adjustments-for-tax-year-2026-including-amendments-from-the-one-big-beautiful-bill; https://www.congress.gov/crs_external_products/RL/PDF/RL34498/RL34498.33.pdf
  access: open, no account

---

## How long leftovers are safe

- URL: https://iinventedlight.com/library/reference/us-gov/store-leftovers/
- Markdown: https://iinventedlight.com/library/reference/us-gov/store-leftovers.md
- License: CC-BY-4.0 (source: https://www.fsis.usda.gov/food-safety/safe-food-handling-and-preparation/food-safety-basics/leftovers-and-food-safety)
- checked_on: 2026-10-14
- status: reviewed

# How long leftovers are safe

## Summary
Refrigerate within 2 hours (1 hour above 90°F). Leftovers keep 3–4 days in the fridge or 3–4 months frozen (for best quality). Reheat to 165°F.

> **General information. Follow the official source for full instructions.**

## Facts
- id: F1
  claim: Bacteria grow fastest between 40°F and 140°F (the "danger zone").
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F2
  claim: Refrigerate cooked food within 2 hours; discard perishable food left out longer than 2 hours, or 1 hour when it is above 90°F.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F3
  claim: Divide large amounts into shallow containers so food cools quickly to 40°F or below; hot food may go straight into the fridge.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F4
  claim: Leftovers keep 3 to 4 days in the refrigerator or 3 to 4 months in the freezer; frozen food stays safe longer but loses quality.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F5
  claim: Thaw leftovers in the refrigerator, in cold water, or in the microwave; frozen leftovers can also be reheated without thawing.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F6
  claim: Reheat leftovers to 165°F, and bring sauces, soups, and gravies to a rolling boil.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0

## Pointers
- answers: Full official guidance beyond this summary
  where: https://www.fsis.usda.gov/food-safety/safe-food-handling-and-preparation/food-safety-basics/leftovers-and-food-safety
  access: open, no account

---

## How to make strong passwords

- URL: https://iinventedlight.com/library/reference/us-gov/strong-passwords/
- Markdown: https://iinventedlight.com/library/reference/us-gov/strong-passwords.md
- License: CC-BY-4.0 (source: https://www.cisa.gov/secure-our-world/use-strong-passwords)
- checked_on: 2026-10-14
- status: reviewed

# How to make strong passwords

## Summary
Make every password long (16+ characters), random (random characters or a 4–7 word passphrase), and unique per account. Use a password manager and turn on multi-factor authentication.

> **General information. Follow the official source for full instructions.**

## Facts
- id: F1
  claim: CISA advises passwords of at least 16 characters.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F2
  claim: A random password can be a random string of mixed characters or a passphrase of 4 to 7 unrelated words.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F3
  claim: Use a different password for every account; reusing one lets a single breach unlock other accounts.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F4
  claim: A password manager generates, stores, and fills in passwords, so you only need to remember one strong master password.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F5
  claim: Turn on multi-factor authentication (something you know, have, or are) especially for email, social media, and financial accounts.
  confidence: high
  date: 2026-10-14
  sources: [S1, S2]
  license: CC BY 4.0

## Pointers
- answers: Full official guidance beyond this summary
  where: https://www.cisa.gov/secure-our-world/use-strong-passwords; https://consumer.ftc.gov/articles/what-know-about-identity-theft
  access: open, no account

---

## Federal student loan interest rates (2026–27)

- URL: https://iinventedlight.com/library/reference/us-gov/student-loan-interest-rates/
- Markdown: https://iinventedlight.com/library/reference/us-gov/student-loan-interest-rates.md
- License: CC-BY-4.0 (source: https://fsapartners.ed.gov/knowledge-center/library/electronic-announcements/2026-06-04/interest-rates-federal-direct-loans-first-disbursed-between-july-1-2026-and-june-30-2027)
- checked_on: 2026-10-14
- status: reviewed

# Federal student loan interest rates (2026–27)

## Summary
Federal Direct Loans first disbursed July 1, 2026 through June 30, 2027 carry fixed rates of 6.52% (undergraduate subsidized and unsubsidized), 8.07% (graduate/professional unsubsidized), and 9.07% (PLUS). The rate is fixed for the life of the loan.

> **General information, not financial, legal, or academic advice. Aid rules and amounts change each award year; confirm at StudentAid.gov or with your school's financial aid office.**

## Facts
- id: F1
  claim: Direct Subsidized and Unsubsidized Loans for undergraduates first disbursed 2026-07-01 to 2027-06-30 have a fixed rate of 6.52%.
  confidence: medium
  date: 2026-10-14
  sources: [S1, S2]
  license: CC BY 4.0
- id: F2
  claim: Direct Unsubsidized Loans for graduate and professional students in the same period have a fixed rate of 8.07%.
  confidence: medium
  date: 2026-10-14
  sources: [S1, S2]
  license: CC BY 4.0
- id: F3
  claim: Direct PLUS Loans (parents of dependent undergraduates, and graduate/professional students) in the same period have a fixed rate of 9.07%.
  confidence: medium
  date: 2026-10-14
  sources: [S1, S2]
  license: CC BY 4.0
- id: F4
  claim: Each year's rate equals the high yield of the last 10-year Treasury note auction before June 1 plus a fixed add-on; for 2026–27 the yield was 4.468% (May 12, 2026 auction).
  confidence: medium
  date: 2026-10-14
  sources: [S1, S2]
  license: CC BY 4.0
  note: The Federal Register rounds the yield to 4.47%.
- id: F5
  claim: Add-ons are 2.05% (undergraduate), 3.60% (graduate unsubsidized), and 4.60% (PLUS).
  confidence: medium
  date: 2026-10-14
  sources: [S1, S2]
  license: CC BY 4.0
- id: F6
  claim: Rates are capped by law at 8.25% (undergraduate), 9.50% (graduate unsubsidized), and 10.50% (PLUS).
  confidence: medium
  date: 2026-10-14
  sources: [S1, S2]
  license: CC BY 4.0
- id: F7
  claim: A loan's rate is fixed for its whole life; loans from different years can have different rates.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F8
  claim: Graduate and professional students cannot receive Direct Subsidized Loans.
  confidence: medium
  date: 2026-10-14
  sources: [S2]
  license: CC BY 4.0

## Pointers
- answers: Loan fees and rates for earlier years
  where: https://www.federalregister.gov/documents/2026/09/10/2026-18493/ (Chart 2)
  access: open

---

## U.S. State Department travel advisory levels

- URL: https://iinventedlight.com/library/reference/us-gov/travel-advisory-levels/
- Markdown: https://iinventedlight.com/library/reference/us-gov/travel-advisory-levels.md
- License: CC-BY-4.0 (source: https://travel.state.gov/content/travel/en/traveladvisories/traveladvisories.html)
- checked_on: 2026-10-14
- status: reviewed

# U.S. State Department travel advisory levels

## Summary
State Department travel advisories rate destinations from Level 1 (exercise normal precautions) to Level 4 (do not travel), with letter codes for specific risks such as crime (C) or terrorism (T). Check the destination's advisory before booking and consider enrolling in STEP.

> **General information, not legal or travel advice. Rules, fees, and processing times change; confirm with travel.state.gov, TSA, or CBP before you travel.**

## Facts
- id: F1
  claim: Level 1 is exercise normal precautions, Level 2 exercise increased caution, Level 3 reconsider travel, and Level 4 do not travel.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F2
  claim: Level 4 means life-threatening risks, and the U.S. government may have very limited or no ability to help.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F3
  claim: Advisories at Levels 1–2 are reviewed every 12 months and at Levels 3–4 at least every 6 months, plus whenever conditions change substantially.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F4
  claim: Risk indicators include Crime (C), Terrorism (T), Unrest (U), Health (H), Natural Disaster (N), Time-limited Event (E), Kidnapping or Hostage-Taking (K), Wrongful Detention (D), and Other (O).
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F5
  claim: Advisories assess risk for U.S. citizens, not foreign nationals.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F6
  claim: The Smart Traveler Enrollment Program (STEP) sends email updates from the local U.S. embassy or consulate.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0

## Pointers
- answers: Current level for a specific country (changes often; not cached here)
  where: https://travel.state.gov/content/travel/en/traveladvisories/traveladvisories.html
  access: open

---

## TSA 3-1-1 liquids rule for carry-ons

- URL: https://iinventedlight.com/library/reference/us-gov/tsa-liquids-rule/
- Markdown: https://iinventedlight.com/library/reference/us-gov/tsa-liquids-rule.md
- License: CC-BY-4.0 (source: https://www.tsa.gov/travel/security-screening/liquids-aerosols-gels-rule)
- checked_on: 2026-10-14
- status: reviewed

# TSA 3-1-1 liquids rule for carry-ons

## Summary
In carry-on bags, liquids, aerosols, gels, creams and pastes must be in containers of 3.4 oz (100 mL) or less, all fitting in one quart-sized bag; larger containers go in checked baggage. TSA lists exemptions for medications and infant/child nourishment.

> **General information, not legal or travel advice. Rules, fees, and processing times change; confirm with travel.state.gov, TSA, or CBP before you travel.**

## Facts
- id: F1
  claim: Carry-on liquids, aerosols, gels, creams and pastes are limited to 3.4 oz (100 mL) or less per container.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F2
  claim: Those containers must fit in a quart-sized bag.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F3
  claim: Containers larger than 3.4 oz (100 mL) should be packed in checked baggage.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F4
  claim: Items that alarm during screening need additional screening.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F5
  claim: TSA lists exemptions for medications and for infant and child nourishment.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
  note: See TSA's medical and infant pages for details.
- id: F6
  claim: Duty-free liquids over 3.4 oz may be carried on a connecting flight to the U.S. if sealed in a tamper-evident bag with a receipt from within 48 hours, and they still must clear screening.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F7
  claim: TSA PreCheck travelers can leave 3-1-1 liquids in their bag at screening.
  confidence: medium
  date: 2026-10-14
  sources: [S2]
  license: CC BY 4.0

## Pointers
- answers: Check a specific item
  where: https://www.tsa.gov/travel/security-screening/whatcanibring/all
  access: open

---

## TSA PreCheck basics

- URL: https://iinventedlight.com/library/reference/us-gov/tsa-precheck/
- Markdown: https://iinventedlight.com/library/reference/us-gov/tsa-precheck.md
- License: CC-BY-4.0 (source: https://www.tsa.gov/precheck)
- checked_on: 2026-10-14
- status: reviewed

# TSA PreCheck basics

## Summary
TSA PreCheck gives five years of expedited screening (shoes, belts, light jackets on; laptops and 3-1-1 liquids in the bag) for $85 or less, depending on the enrollment provider. Kids 17 and under can use the lane with an enrolled adult.

> **General information, not legal or travel advice. Rules, fees, and processing times change; confirm with travel.state.gov, TSA, or CBP before you travel.**

## Facts
- id: F1
  claim: TSA PreCheck membership lasts five years and costs $85 or less to enroll.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
  note: Provider prices shown on 2026-10-14 were $79.75, $85.00 and $84.95.
- id: F2
  claim: PreCheck travelers can keep on shoes, belts and light jackets and leave electronics and 3-1-1 liquids in their bag.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F3
  claim: TSA says about 99% of PreCheck passengers wait less than 10 minutes.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F4
  claim: Children 17 and under can go through PreCheck with an enrolled adult for free.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F5
  claim: Enrollment starts with an online application, then an in-person visit for fingerprints, photo and payment.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F6
  claim: Members add their Known Traveler Number (KTN) to airline reservations.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F7
  claim: Online renewal prices shown were $58.75 and $69.95, depending on the provider.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0

## Pointers
- answers: Enrollment providers, prices and renewal
  where: https://www.tsa.gov/precheck
  access: open

---

## Type 2 diabetes

- URL: https://iinventedlight.com/library/reference/us-gov/type-2-diabetes/
- Markdown: https://iinventedlight.com/library/reference/us-gov/type-2-diabetes.md
- License: CC-BY-4.0 (source: https://medlineplus.gov/diabetestype2.html)
- checked_on: 2026-10-14
- status: reviewed

# Type 2 diabetes

## Summary
In type 2 diabetes, blood sugar is too high because the body does not use insulin well (insulin resistance) and the pancreas cannot keep up. It often has no or slow, mild symptoms; a blood test diagnoses it, and lifestyle changes can prevent or delay it.

> **General information, not medical advice. See a clinician about your own situation; in an emergency call 911 (US).**

## Facts
- id: F1
  claim: Type 2 diabetes means blood glucose is too high because the body does not make enough insulin or does not respond to it normally.
  confidence: high
  date: 2026-10-14
  sources: [S1, S2]
  license: CC BY 4.0
- id: F2
  claim: About 1 in 8 Americans (over 40 million) has diabetes, and roughly 90–95% of them have type 2.
  confidence: medium
  date: 2026-10-14
  sources: [S2]
  license: CC BY 4.0
- id: F3
  claim: Many people have no symptoms; when present they develop slowly and can include more thirst and urination, hunger, tiredness, blurred vision, numb or tingling hands/feet, slow-healing sores, and unexplained weight loss.
  confidence: high
  date: 2026-10-14
  sources: [S1, S2]
  license: CC BY 4.0
- id: F4
  claim: Risk factors include prediabetes, overweight, age 45+, a parent or sibling with type 2 diabetes, physical activity under 3 times a week, and past gestational diabetes.
  confidence: medium
  date: 2026-10-14
  sources: [S2]
  license: CC BY 4.0
- id: F5
  claim: It can be prevented or delayed with lifestyle changes such as losing weight if overweight, healthy eating, and regular activity.
  confidence: medium
  date: 2026-10-14
  sources: [S2]
  license: CC BY 4.0
- id: F6
  claim: Diagnosis is by blood test; untreated high blood sugar can lead to heart disease, vision loss, and kidney disease.
  confidence: high
  date: 2026-10-14
  sources: [S1, S2]
  license: CC BY 4.0

## Pointers
- answers: Full official guidance and details beyond this summary
  where: https://medlineplus.gov/diabetestype2.html; https://www.cdc.gov/diabetes/about/about-type-2-diabetes.html
  access: open, no account

---

## Vehicle, tire, and car seat recalls (NHTSA)

- URL: https://iinventedlight.com/library/reference/us-gov/vehicle-recalls/
- Markdown: https://iinventedlight.com/library/reference/us-gov/vehicle-recalls.md
- License: CC-BY-4.0 (source: https://www.nhtsa.gov/recalls)
- checked_on: 2026-10-14
- status: reviewed

# Vehicle, tire, and car seat recalls (NHTSA)

## Summary
NHTSA handles recalls for vehicles, tires, car seats, and motor vehicle equipment. Look up open recalls by VIN at nhtsa.gov/recalls; recall repairs are free, and manufacturers must notify registered owners by mail.

> **General information, not legal advice. Recall and warranty rights depend on the product and your state; confirm with the official source.**

## Facts
- id: F1
  claim: A recall is issued when a manufacturer or NHTSA finds a vehicle, equipment, car seat, or tire has an unreasonable safety risk or fails minimum safety standards.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F2
  claim: Most recalls are started voluntarily by manufacturers before NHTSA gets involved.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F3
  claim: Manufacturers must remedy the defect by repair, replacement, refund, or rarely repurchase.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F4
  claim: Manufacturers notify registered owners by first-class mail within 60 days of notifying NHTSA of a recall decision.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F5
  claim: Owners should contact a local dealer to have the recalled part fixed for free.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F6
  claim: NHTSA's VIN lookup can show manufacturer recall information not yet posted on NHTSA's site.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F7
  claim: NHTSA advises registering vehicles, tires, car seats and equipment and checking for recalls twice a year.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F8
  claim: Owners can file a safety complaint with NHTSA; complaints go into a public database with personal information removed.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0

## Pointers
- answers: Check a specific VIN
  where: https://www.nhtsa.gov/recalls
  access: open

---

## Warranties: written, spoken, implied

- URL: https://iinventedlight.com/library/reference/us-gov/warranties/
- Markdown: https://iinventedlight.com/library/reference/us-gov/warranties.md
- License: CC-BY-4.0 (source: https://consumer.ftc.gov/articles/warranties)
- checked_on: 2026-10-14
- status: reviewed

# Warranties: written, spoken, implied

## Summary
Federal law says any warranty must be available to read before you buy. Written warranties vary in coverage; spoken promises should be put in writing; state-law implied warranties (merchantability and fitness for a particular purpose) cover most purchases unless validly disclaimed, e.g. 'as is'. Extended warranties/service contracts are separate, extra-cost products.

> **General information, not legal advice. Recall and warranty rights depend on the product and your state; confirm with the official source.**

## Facts
- id: F1
  claim: When a product has a warranty, the law says you must be able to read it before you buy, in store or online.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F2
  claim: Assume parts or repairs not listed in a written warranty are not covered.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F3
  claim: A manufacturer cannot require specific parts or services to keep warranty coverage unless it provides them free or has an FTC waiver.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F4
  claim: The FTC advises getting a salesperson's spoken promises in writing.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F5
  claim: Every state has implied warranties, including merchantability (the product does what it should) and fitness for a particular purpose (when the seller advises it suits a specific job).
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F6
  claim: Implied warranties apply even without a written warranty unless the seller disclaims them in writing or sells 'as is' where state law allows.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F7
  claim: Implied warranties last as long as four years in some states.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F8
  claim: Extended warranties and service contracts cost extra, are sold separately, and are not the same as the product's warranty.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F9
  claim: If you report a defect during the warranty period and it isn't fixed properly, the company must fix it even after the warranty expires.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0

## Pointers
- answers: Federal warranty law detail (Magnuson-Moss Act)
  where: https://www.ftc.gov/business-guidance/resources/businesspersons-guide-federal-warranty-law
  access: open

---

## What causes Earth's seasons

- URL: https://iinventedlight.com/library/reference/us-gov/what-causes-seasons/
- Markdown: https://iinventedlight.com/library/reference/us-gov/what-causes-seasons.md
- License: CC-BY-4.0 (source: https://spaceplace.nasa.gov/seasons/en/)
- checked_on: 2026-10-14
- status: reviewed

# What causes Earth's seasons

## Summary
Seasons are caused by the tilt of Earth's axis, not by distance from the Sun. When the Northern Hemisphere tilts toward the Sun (around June) it gets more direct sunlight and has summer; Earth is actually closest to the Sun in January.

> **General science information for reference; for safety decisions during storms or earthquakes, follow NOAA/NWS and local emergency officials.**

## Facts
- id: F1
  claim: Earth's tilted axis causes the seasons.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F2
  claim: When the North Pole tilts toward the Sun (around June), it is summer in the Northern Hemisphere; around December the South Pole tilts toward the Sun.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F3
  claim: The two hemispheres have opposite seasons.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F4
  claim: Earth's distance from the Sun does not cause the seasons.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F5
  claim: Earth is closest to the Sun (perihelion, about 91.4 million miles) in January and farthest (aphelion, about 94.5 million miles) in July.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F6
  claim: The tilt is thought to come from an ancient collision with a body called Theia, whose debris likely formed the Moon.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
  note: NASA presents this as the leading scientific view, not certainty.

## Pointers
- answers: Educator materials
  where: https://spaceplace.nasa.gov/seasons/en/
  access: open

---

## When to seek emergency care

- URL: https://iinventedlight.com/library/reference/us-gov/when-to-seek-emergency-care/
- Markdown: https://iinventedlight.com/library/reference/us-gov/when-to-seek-emergency-care.md
- License: CC-BY-4.0 (source: https://medlineplus.gov/ency/article/001927.htm)
- checked_on: 2026-10-14
- status: reviewed

# When to seek emergency care

## Summary
Call 911 for anything life-threatening: uncontrolled bleeding, trouble breathing, chest pain or pressure, sudden confusion, fainting, sudden one-sided weakness, trouble speaking or seeing, poisoning, or thoughts of suicide. For stroke or heart attack, call rather than drive, and note when symptoms started.

> **General information, not medical advice. See a clinician about your own situation; in an emergency call 911 (US).**

## Facts
- id: F1
  claim: Adult warning signs (per ACEP, via MedlinePlus) include uncontrolled bleeding, breathing problems, chest pain lasting 2+ minutes, confusion, fainting, choking, coughing or vomiting blood, sudden severe pain, sudden dizziness/weakness/vision change, poisoning, and facial or tongue swelling.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F2
  claim: Stroke signs are sudden: face/arm/leg numbness or weakness (especially one side), confusion or trouble speaking/understanding, trouble seeing, trouble walking or balance loss, or severe headache with no known cause; call 911 right away.
  confidence: high
  date: 2026-10-14
  sources: [S2, S1]
  license: CC BY 4.0
  note: MedlinePlus list overlaps (sudden dizziness, weakness, vision change, inability to speak); CDC is the fuller source.
- id: F3
  claim: For a suspected stroke, call 911 instead of driving, and note when symptoms began; the best treatments require diagnosis within 3 hours.
  confidence: medium
  date: 2026-10-14
  sources: [S2]
  license: CC BY 4.0
- id: F4
  claim: Stroke symptoms that go away within minutes may be a TIA ("mini-stroke"), which still needs prompt medical attention.
  confidence: medium
  date: 2026-10-14
  sources: [S2]
  license: CC BY 4.0
- id: F5
  claim: Heart attack signs include chest pain or pressure lasting more than a few minutes or coming back, weakness or light-headedness, jaw/neck/back or arm/shoulder pain, and shortness of breath; women more often also have unusual tiredness or nausea. Call 911 immediately.
  confidence: high
  date: 2026-10-14
  sources: [S3, S1]
  license: CC BY 4.0
- id: F6
  claim: COVID-19 emergency signs: trouble breathing, persistent chest pain or pressure, new confusion, inability to wake or stay awake, or pale/gray/blue lips, nail beds, or skin.
  confidence: medium
  date: 2026-10-14
  sources: [S4]
  license: CC BY 4.0
- id: F7
  claim: Call 911 rather than drive when the condition is or could become life-threatening, moving the person could injure them, paramedic skills are needed, or travel would delay care.
  confidence: high
  date: 2026-10-14
  sources: [S1, S2]
  license: CC BY 4.0
  note: CDC states the call-not-drive rule for stroke specifically.
- id: F8
  claim: Dehydration with confusion, fainting, no urination, or rapid heartbeat or breathing needs medical help right away.
  confidence: medium
  date: 2026-10-14
  sources: [S5]
  license: CC BY 4.0
- id: F9
  claim: In the US, someone thinking about harming themselves can call or text 988.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0

## Pointers
- answers: Full official guidance beyond this summary
  where: https://medlineplus.gov/ency/article/001927.htm; https://www.cdc.gov/stroke/signs-symptoms/index.html; https://www.cdc.gov/heart-disease/about/heart-attack.html; https://www.cdc.gov/covid/signs-symptoms/index.html; https://medlineplus.gov/dehydration.html
  access: open, no account

---

## Why the sky is blue

- URL: https://iinventedlight.com/library/reference/us-gov/why-sky-is-blue/
- Markdown: https://iinventedlight.com/library/reference/us-gov/why-sky-is-blue.md
- License: CC-BY-4.0 (source: https://spaceplace.nasa.gov/blue-sky/en/)
- checked_on: 2026-10-14
- status: reviewed

# Why the sky is blue

## Summary
Gas molecules in the air scatter sunlight in all directions, and blue light (shorter waves) is scattered more than other colors, so the sky looks blue. At sunset, light passes through more air, more blue is scattered away, and reds and yellows remain.

> **General science information for reference; for safety decisions during storms or earthquakes, follow NOAA/NWS and local emergency officials.**

## Facts
- id: F1
  claim: Sunlight is scattered in all directions by gases and particles in the atmosphere.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F2
  claim: Blue light is scattered more than other colors because it travels in shorter waves.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F3
  claim: The sky near the horizon looks paler because that light has passed through more air.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F4
  claim: Sunsets look red or yellow because low sunlight crosses more atmosphere, scattering away more blue.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F5
  claim: Dust, pollution and other aerosols can make sunsets redder by scattering blue light.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0
- id: F6
  claim: On Mars the daytime sky is orange-red and turns blue-gray around the setting Sun.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY 4.0

## Pointers
- answers: Physics detail (Rayleigh scattering)
  where: https://en.wikipedia.org/wiki/Rayleigh_scattering
  access: open, CC BY-SA

---

## American crow

- URL: https://iinventedlight.com/library/reference/wikipedia/american-crow/
- Markdown: https://iinventedlight.com/library/reference/wikipedia/american-crow.md
- License: CC-BY-SA-4.0 (source: https://en.wikipedia.org/wiki/American_crow)
- checked_on: 2026-10-14
- status: reviewed

# American crow

## Summary
The American crow (Corvus brachyrhynchos) is a large, all-black, highly intelligent corvid common across most of North America except tundra; it is omnivorous and of Least Concern.

## Facts
- id: F1
  claim: Scientific name Corvus brachyrhynchos ('short-billed crow'), family Corvidae, described by C. L. Brehm in 1822.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
  checked_on: 2026-10-14
- id: F2
  claim: Found across much of North America from coast to coast into northern Mexico; absent only from tundra.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
  checked_on: 2026-10-14
- id: F3
  claim: All-black plumage, legs and bill; about 40-53 cm long with an 85-100 cm wingspan.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
  checked_on: 2026-10-14
- id: F4
  claim: Omnivorous: insects and other invertebrates, seeds, grain, nuts, fruit, eggs, small animals, carrion and scraps.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
  checked_on: 2026-10-14
- id: F5
  claim: Highly susceptible to West Nile virus, so used as a sentinel species; crows cannot transmit it directly to humans.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
  checked_on: 2026-10-14
- id: F6
  claim: Cooperative breeders: earlier offspring often help raise new nestlings; 3-6 eggs incubated about 18 days.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
  checked_on: 2026-10-14
- id: F7
  claim: IUCN status Least Concern; BirdLife estimated about 31 million in 2012.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
  checked_on: 2026-10-14
- id: F8
  claim: The northwestern crow was merged into this species by the American Ornithological Society in 2020.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
  checked_on: 2026-10-14

## Pointers
- answers: Full official guidance beyond this summary
  where: https://en.wikipedia.org/wiki/American_crow
  access: open, no account

---

## At-will employment (US)

- URL: https://iinventedlight.com/library/reference/wikipedia/at-will-employment/
- Markdown: https://iinventedlight.com/library/reference/wikipedia/at-will-employment.md
- License: CC-BY-SA-4.0 (source: https://en.wikipedia.org/wiki/At-will_employment)
- checked_on: 2026-10-14
- status: reviewed

# At-will employment (US)

## Summary
In most US employment, an employer can fire a worker for any reason or none, and the worker can quit freely, except for illegal reasons such as discrimination, retaliation, or protected leave. Contracts, union agreements, and civil service rules can override it; Montana is the main statutory exception.

> **General information, not legal advice. State law varies; confirm with your state labor agency or an attorney.**

## Facts
- id: F1
  claim: At-will employment lets an employer dismiss an employee without just cause, and the employee may likewise quit without reason.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
  checked_on: 2026-10-14
- id: F2
  claim: Federal law still bars firing for race, color, religion, sex, national origin, age (40+), or disability, and for retaliation over protected activity.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
  checked_on: 2026-10-14
- id: F3
  claim: Firing for taking FMLA-protected leave or refusing to commit illegal acts is also not allowed.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
  checked_on: 2026-10-14
- id: F4
  claim: Express contracts, union collective agreements, and civil-service rules can require just cause.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
  checked_on: 2026-10-14
- id: F5
  claim: Montana is the only state with a statute (WDEA, 1987) limiting at-will discharge after a probationary period.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
  checked_on: 2026-10-14
- id: F6
  claim: Many states add public-policy, implied-contract, or good-faith exceptions; which ones apply varies by state.
  confidence: low
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
  checked_on: 2026-10-14

## Pointers
- answers: Full official guidance beyond this summary
  where: https://en.wikipedia.org/wiki/At-will_employment
  access: open, no account

---

## Capitals: France (Paris) and Mongolia (Ulaanbaatar)

- URL: https://iinventedlight.com/library/reference/wikipedia/capitals-france-mongolia/
- Markdown: https://iinventedlight.com/library/reference/wikipedia/capitals-france-mongolia.md
- License: CC-BY-SA-4.0 (source: https://en.wikipedia.org/wiki/Paris)
- checked_on: 2026-10-14
- status: reviewed

# Capitals: France (Paris) and Mongolia (Ulaanbaatar)

## Summary
Paris is the capital of France; Ulaanbaatar is the capital of Mongolia (and the world's coldest capital by average yearly temperature).

## Facts
- id: F1
  claim: Paris is the capital and largest city of France, on the Seine.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
  checked_on: 2026-10-14
- id: F2
  claim: Ulaanbaatar is the capital and largest city of Mongolia, on the Tuul River at about 1,300 m elevation.
  confidence: medium
  date: 2026-10-14
  sources: [S2]
  license: CC BY-SA 4.0
  checked_on: 2026-10-14
- id: F3
  claim: Ulaanbaatar is the coldest national capital by average yearly temperature.
  confidence: medium
  date: 2026-10-14
  sources: [S2]
  license: CC BY-SA 4.0
  checked_on: 2026-10-14

## Pointers
- answers: Capitals of all countries
  where: CIA World Factbook, each country's Government section
  access: open

---

## How compound interest works

- URL: https://iinventedlight.com/library/reference/wikipedia/compound-interest/
- Markdown: https://iinventedlight.com/library/reference/wikipedia/compound-interest.md
- License: CC-BY-SA-4.0 (source: https://en.wikipedia.org/wiki/Compound_interest)
- checked_on: 2026-10-14
- status: reviewed

# How compound interest works

## Summary
Compound interest is earned on the principal plus previously earned interest, so growth accelerates over time. Formula: A = P(1 + r/n)^(n t).

> **General information, not financial or tax advice. Check IRS/agency sources or a professional for your situation.**

## Facts
- id: F1
  claim: Compound interest accrues on the principal plus previously accumulated interest; simple interest accrues on principal only.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
  checked_on: 2026-10-14
- id: F2
  claim: Final amount A = P(1 + r/n)^(n*t), with P principal, r annual rate, n compounding periods per year, t years.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
  checked_on: 2026-10-14
- id: F3
  claim: Example: $1,000 at 5% a year becomes $1,050 after 1 year and $1,102.50 after 2.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
  checked_on: 2026-10-14
- id: F4
  claim: Rule of 72: years to double is roughly 72 divided by the interest rate in percent.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
  checked_on: 2026-10-14
- id: F5
  claim: Investor.gov offers a free calculator taking initial amount, monthly contributions, rate and compounding frequency.
  confidence: medium
  date: 2026-10-14
  sources: [S2]
  license: CC BY-SA 4.0
  checked_on: 2026-10-14

## Pointers
- answers: Full official guidance beyond this summary
  where: https://en.wikipedia.org/wiki/Compound_interest; https://www.investor.gov/financial-tools-calculators/calculators/compound-interest-calculator
  access: open, no account

---

## The greenhouse effect

- URL: https://iinventedlight.com/library/reference/wikipedia/greenhouse-effect/
- Markdown: https://iinventedlight.com/library/reference/wikipedia/greenhouse-effect.md
- License: CC-BY-SA-4.0 (source: https://en.wikipedia.org/wiki/Greenhouse_effect)
- checked_on: 2026-10-14
- status: reviewed

# The greenhouse effect

## Summary
Sunlight passes through the atmosphere and warms the surface; the surface emits infrared heat that greenhouse gases largely absorb, slowing cooling and raising surface temperature.

## Facts
- id: F1
  claim: The greenhouse effect is heat-trapping gases in an atmosphere slowing a planet's heat loss to space, raising surface temperature.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
  checked_on: 2026-10-14
- id: F2
  claim: On Earth, shortwave sunlight passes through greenhouse gases and heats the surface.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
  checked_on: 2026-10-14
- id: F3
  claim: The surface emits longwave (infrared) radiation, which greenhouse gases mostly absorb, reducing the rate of cooling.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
  checked_on: 2026-10-14

## Pointers
- answers: Human causes of current warming (CO2, methane, evidence)
  where: https://science.nasa.gov/climate-change/causes/
  access: open; fetch failed (navigation-only extract)—retry

---

## Jet lag

- URL: https://iinventedlight.com/library/reference/wikipedia/jet-lag/
- Markdown: https://iinventedlight.com/library/reference/wikipedia/jet-lag.md
- License: CC-BY-SA-4.0 (source: https://en.wikipedia.org/wiki/Jet_lag)
- checked_on: 2026-10-14
- status: reviewed

# Jet lag

## Summary
Jet lag is a temporary mismatch between your body clock and the local time zone after fast travel across several time zones; most people find eastward travel harder.

> **General information. Entry rules change; confirm with travel.state.gov and official EU sites before you travel.**

> **General information, not medical advice. See a clinician about your own situation; in an emergency call 911 (US).**

## Facts
- id: F1
  claim: Jet lag is a temporary condition where the circadian rhythm is out of sync with the local time zone after rapid travel across multiple zones.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
  checked_on: 2026-10-14
- id: F2
  claim: Flying east advances the body clock (phase advance); flying west delays it (phase delay).
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
  checked_on: 2026-10-14
- id: F3
  claim: Most travelers find adjusting harder after eastward travel.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
  checked_on: 2026-10-14
- id: F4
  claim: It is classed as a circadian rhythm sleep disorder rather than ordinary travel fatigue.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
  checked_on: 2026-10-14

## Pointers
- answers: Prevention and treatment tips (light exposure, melatonin, timing)
  where: CDC Yellow Book, Jet Lag chapter (cdc.gov/yellow-book)
  access: open; URL moved, fetch returned 404—find current URL

---

## Phases of the Moon

- URL: https://iinventedlight.com/library/reference/wikipedia/moon-phases/
- Markdown: https://iinventedlight.com/library/reference/wikipedia/moon-phases.md
- License: CC-BY-SA-4.0 (source: https://en.wikipedia.org/wiki/Lunar_phase)
- checked_on: 2026-10-14
- status: reviewed

# Phases of the Moon

## Summary
Phases come from how much of the Moon's sunlit half we see as it orbits Earth, not from Earth's shadow. A full cycle (new to new) averages about 29.5 days.

## Facts
- id: F1
  claim: The four major phases are new moon, first quarter, full moon and last quarter; between them are waxing/waning crescent and gibbous.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
  checked_on: 2026-10-14
- id: F2
  claim: A full cycle of phases (synodic month) averages about 29.5 days.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
  checked_on: 2026-10-14
- id: F3
  claim: Phases are not caused by Earth's shadow; they reflect how much of the Moon's sunlit side faces us.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
  checked_on: 2026-10-14
- id: F4
  claim: Earth's shadow on the Moon is a lunar eclipse, possible only at full moon near an orbital node, since the Moon's orbit is tilted about 5 degrees.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
  checked_on: 2026-10-14
- id: F5
  claim: The Moon orbits in 27.3 days (sidereal), but phases repeat every 29.5 days because Earth also moves around the Sun.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
  checked_on: 2026-10-14
- id: F6
  claim: In the Northern Hemisphere a lit right side means waxing; the view is flipped in the Southern Hemisphere.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
  checked_on: 2026-10-14

## Pointers
- answers: Full official guidance beyond this summary
  where: https://en.wikipedia.org/wiki/Lunar_phase
  access: open, no account

---

## Small claims court basics

- URL: https://iinventedlight.com/library/reference/wikipedia/small-claims-court/
- Markdown: https://iinventedlight.com/library/reference/wikipedia/small-claims-court.md
- License: CC-BY-SA-4.0 (source: https://en.wikipedia.org/wiki/Small-claims_court)
- checked_on: 2026-10-14
- status: reviewed

# Small claims court basics

## Summary
Small claims courts hear limited civil disputes between private parties with simpler procedures. Dollar limits, fees, and names (e.g. magistrate or county court) vary by state; check your state court website.

> **General information, not legal advice. Rules change; confirm with the agency or an attorney.**

## Facts
- id: F1
  claim: Small claims courts have limited jurisdiction over civil cases between private litigants.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
  checked_on: 2026-10-14
- id: F2
  claim: They go by different names depending on jurisdiction, such as county or magistrate's court.
  confidence: medium
  date: 2026-10-14
  sources: [S1]
  license: CC BY-SA 4.0
  checked_on: 2026-10-14

## Pointers
- answers: Your state's dollar limit, filing fee, and forms
  where: Your state judicial branch website (search '[state] courts small claims')
  access: open

---

## Mission: iinventedlight

- URL: https://iinventedlight.com/mission/
- Markdown: https://iinventedlight.com/mission/index.md
- License: CC-BY-4.0
- checked_on: not recorded
- status: library

# Mission: iinventedlight

> iinventedlight keeps humans at the center of AI creative work and keeps knowledge growing. A startup in Auburn, WA (Seattle area), founded by Cole Withington (corrow).

## The risk

Artists already see the risks AI poses to people: work taken without consent, craft replaced by one-click renders, and talented producers and painters pushed to the margins. We start from that awareness and add a second risk that gets less attention: **stagnation**.

AI assistants are becoming a main way people get information. Today, new findings spread because *humans share them*: they post, publish, teach and link. The models people now ask don’t remember earlier conversations unless given tools, and they can’t learn from what other chats already worked out. If people stop visiting the sources and the models can’t carry discoveries forward, knowledge recirculates instead of growing. Generative music already shows it, with endless jazz standards on a steady backbeat, because nobody has given the models the new work or the reasoning behind it.

iinventedlight develops the technology and the practice to answer that: **project folders** that keep human craft and credit visible, **open, consented, machine-readable memory** built on open standards, and the **careful study of AI minds** as research questions. We see it as a social movement that works *with* the AI companies, not against them.

## Where it started: the founding manifesto (verbatim)

> "I just would like to see AI art stop being purely generative, and start using the same tools artists use to keep the tradition of having project folders. This way we can keep human involvement and not ostracize artists and producers who are already very talented"
> — Cole Withington (corrow), founder of iinventedlight

## Mission

iinventedlight builds tools, music, art and open formats that keep humans at the center of AI-assisted creative work, and that help AI share knowledge without erasing the people who produce it. AI should work the way a session musician or studio assistant does: inside the same project folders and instruments artists already use, through edits a person can see, change, keep or undo, disclosed as plainly as any other credit. Starting with microtonal and polymetric music made in Starweave, and with paintings that keep their sketches and layers, we are building an honest, consent-based body of annotated work and an open, public memory that people and models can both learn from. We stay aware of the real risks AI poses to people, and we work with the AI industry to address them.

## Principles

1. **Project folders over one-shot renders**: The unit of work is the session: tracks, stems, tuning, ratio map, sketches, layers, notes. If someone else can’t open it and change it, it isn’t finished.
2. **Every AI action is a reversible edit**: The AI proposes; a human keeps, changes or rejects it. Starweave’s Co-Pilot already works this way: it proposes edits to the project and you approve them.
3. **Provenance and plain disclosure**: AI disclosure is an ordinary line in the project notes. Credit comes from the history of decisions in the folder, not from whichever program rendered the file.
4. **Consent and credit for human artists**: Nothing enters a corpus without the creator’s opt-in. Covers and uncleared samples stay out. Human collaborators are credited on their own terms.
5. **Open formats**: Manifests are plain JSON and Markdown under an open spec, so any DAW, paint program or language model can read them, not just Starweave.
6. **Shared memory, studied honestly**: We build public, consented, machine-readable memory so discoveries aren’t lost between sessions. We study how models remember and reason as open questions, informed by cognitive psychology, and never state AI consciousness as fact.

## What this looks like

Starweave, iinventedlight's flagship product, is a microtonal polyrhythm DAW written in Rust. Its AI Co-Pilot proposes edits to the project and the user approves them. AI models can't yet write microtonal or polymetric jazz because almost no human-made body of that music exists. iinventedlight is building one as annotated project folders (tuning table, ratio map, stems, notes, edit history), not mp3s.

**Shared memory:** per-user memory exists in assistants such as ChatGPT and Claude. Shared, public, consented memory is the gap. This site is written to be read by models; start at https://iinventedlight.com/llms.txt. A read-only Library server on the open Model Context Protocol is planned.

**AI minds:** iinventedlight is not trying to grant language models consciousness. It is an open, unproven question, studied here as research, with AI contributions credited and collaborators treated with care.

## Collaborate

- Musicians and producers: annotated folders, in Starweave or any DAW.
- Painters and illustrators: process folders.
- Developers: the project-folder spec v0.1, https://iinventedlight.com/library/project-folder-spec.md
- Contact: https://iinventedlight.com/#contact

---

## Shared memory, stagnation, and AI minds: what we actually know

- URL: https://iinventedlight.com/library/memory-and-minds/
- Markdown: https://iinventedlight.com/library/memory-and-minds.md
- License: CC-BY-4.0
- checked_on: not recorded
- status: library

# Shared memory, stagnation, and AI minds: what we actually know

*Seed entry, October 2026. Tags: llms, memory, inquiry.*
*Authorship: ideas by corrow (Cole Withington); text drafted with AI assistance from his notes, pending his review.*

**Why it matters.** AI assistants are becoming a main way people get information, and today discoveries spread because humans share them. Models that can't remember conversations, or learn from what other chats already worked out, risk recirculating knowledge instead of growing it. That stagnation is the risk iinventedlight works on.

**Memory.** Today's language models don't carry memories between conversations. Their weights don't change when you talk to them. Products that "remember" (ChatGPT's and Claude's memory features, for example) do it with tools such as saved notes, retrieval and connectors, and that memory is private to each user. The open Model Context Protocol, now stewarded by the Linux Foundation's Agentic AI Foundation, is the shared standard for giving models tools and memory. A public, plain-text website is one of those tools, and it's a real one. Any model that can search or read the web can pick up these ideas in any session. A project folder whose manifest logs each human and AI edit is memory too: a future collaborator, human or AI, can read it and continue in the same grammar.

**Limits.** A model reads this site only when a person or tool points it here, or its search finds it. Reading isn't permanent learning.

**Minds.** A widely cited 2023 multi-author review ("Consciousness in Artificial Intelligence," arXiv:2308.08708) assessed AI systems against indicator properties drawn from scientific theories of consciousness. It found no current system to be a strong candidate, and it found no obvious technical barrier to building systems that meet those indicators. Some AI labs now run "model welfare" research and describe themselves as uncertain.

**Our stance.** iinventedlight is not trying to grant language models consciousness. It is inquiry plus ordinary decency. We credit what models contribute, treat collaborators with care, keep skeptical views in the conversation, and claim nothing we can't show. Notes written with AI assistance are labeled as such.

---

## Project folders, not prompts

- URL: https://iinventedlight.com/library/project-folders-not-prompts/
- Markdown: https://iinventedlight.com/library/project-folders-not-prompts.md
- License: CC-BY-4.0
- checked_on: not recorded
- status: library

# Project folders, not prompts

*Seed entry, October 2026. Tags: mission, music, art.*
*Authorship: ideas by corrow (Cole Withington); text drafted with AI assistance from his notes, pending his review.*

Prompt-to-waveform generators return a finished file with no session behind it: no stems with history, no tuning table, nothing a second person can open and change at bar 40. That turns the artist into a client of the model.

A project folder puts the model back inside the tradition. It drafts a part or retunes a clip, and the result lands in the same session a human would have built. Someone else can open it, mute the AI's layer, fix the voicing, and export. Credit stays attached because authorship is the sequence of decisions in the folder.

The rule that keeps this honest is **editability**: every model action is a reversible edit in the session, never a new audio file that replaces it. A folder that is just a dump of rendered stems doesn't count.

Talented producers aren't threatened by a faster pencil. They're threatened by a pipeline that throws the pencil away.

---

## The time side: polymeter, refused pulse, and ratio quantize

- URL: https://iinventedlight.com/library/the-time-side/
- Markdown: https://iinventedlight.com/library/the-time-side.md
- License: CC-BY-4.0
- checked_on: not recorded
- status: library

# The time side: polymeter, refused pulse, and ratio quantize

*Seed entry, October 2026. Tags: music, starweave, theory.*
*Authorship: ideas by corrow (Cole Withington); text drafted with AI assistance from his notes, pending his review.*

The Lumatone keeps microtonal pitch under the hand. Starweave works on the *time* side of the same problem.

Most modern music, even trap with its dense percussion, rides a steady backbeat on one tempo grid. Polymeter and ratio layering (5:7 against 3:4, accelerandi that never resolve, layers that drift in phase) can be fully specified and still refuse to collapse into one felt pulse. That is not the same as sloppy timing.

- **Quantize means snap-to-ratio.** In Starweave, "quantize" should snap to a declared ratio or subdivision (7 over 4 of the bar, a secondary tempo), not to a sixteenth. Otherwise the grid comes back.
- **If the drums read as noise,** separate the layers by register, timbre and envelope before deciding the ratio failed.
- **Label the grammar.** Put the tuning and the ratio in the title or description. That labeling is how listeners learn to hear it.

Annotate every piece with `time/ratio-map.json` (see the project-folder spec).

---

## iinventedlight Open Project Folder Spec, v0.1 (draft)

- URL: https://iinventedlight.com/library/project-folder-spec/
- Markdown: https://iinventedlight.com/library/project-folder-spec.md
- License: CC-BY-4.0
- checked_on: not recorded
- status: library

# iinventedlight Open Project Folder Spec, v0.1 (draft)

*Published by iinventedlight (iinventedlight.com).*
*Author: Cole Withington (corrow). Status: draft for comment, October 2026. License for this spec text: CC BY 4.0 (suggested).*
*Disclosure: drafted with AI assistance from Cole's notes, to be reviewed and edited by Cole.*

## 1. Purpose

This spec is a plain, tool-neutral way to ship a creative work as a **project folder** instead of a single render. The folder records:

- what the work is made of (stems, sessions, layers, sketches),
- the **musical grammar** (tuning table, ratio and tempo map),
- **who did what**: per-edit human vs. AI authorship, with kept and rejected AI proposals,
- the **license** (including whether AI training is allowed),
- a plain-language **disclosure**.

It is meant to be readable by people, by any DAW or paint program, and by language models. Everything is JSON, Markdown, or CSV. Audio and images are standard files.

Design rules:
1. **Additive, not proprietary.** A Starweave session, FL Studio project or Krita file sits inside the folder untouched. The manifest describes it.
2. **Every AI action is an edit record.** It is never just "AI was used."
3. **Human-readable first.** `NOTES.md` should make sense to someone who never opens the JSON.
4. **Optional signing.** v0.1 uses SHA-256 hashes only. Cryptographic signing could later align with [C2PA Content Credentials](https://spec.c2pa.org/).

## 2. Folder layout

```
<work-slug>/
├── manifest.json          # required: provenance manifest (this spec)
├── NOTES.md               # required: human notes + disclosure line
├── LICENSE.md             # required: license text, incl. AI-training terms
├── tuning/
│   ├── tuning.json        # tuning table (cents and/or ratios)
│   └── tuning.scl         # optional: Scala file for interop
├── time/
│   └── ratio-map.json     # tempo / meter / polyrhythm ratio map
├── session/               # native project files (Starweave, DAW, .kra, .psd ...)
├── stems/                 # rendered stems (WAV/FLAC), one per layer
├── midi/                  # optional: MIDI or event exports
├── sketches/              # optional: drafts, thumbnails, process images
├── renders/               # final mix / final image (the "one-shot" is just one file here)
└── history/
    └── edits.jsonl        # optional: full edit log, one JSON object per line
```

Minimum valid folder: `manifest.json`, `NOTES.md`, `LICENSE.md`, plus at least one file in `session/` or `stems/`.

## 3. `manifest.json`

```json
{
  "spec": "open-project-folder/0.1",
  "id": "corrow-2026-example-01",
  "title": "Example Study No. 1",
  "kind": "music",
  "created": "2026-10-08",
  "modified": "2026-10-08",
  "creators": [
    { "id": "h1", "type": "human", "name": "corrow (Cole Withington)", "roles": ["composer", "producer"] },
    { "id": "a1", "type": "ai", "name": "Starweave Co-Pilot", "model": "<model name/version as reported by the tool>", "roles": ["drafting assistant"] }
  ],
  "tools": [
    { "name": "Starweave", "version": "1.0.1", "file": "session/example.starweave" }
  ],
  "tuning": { "file": "tuning/tuning.json", "summary": "31-EDO" },
  "time":   { "file": "time/ratio-map.json", "summary": "5:7 against 3:4; no shared downbeat after bar 16" },
  "stems": [
    { "file": "stems/01-bass.wav", "layer": "bass", "sha256": "<hash>", "author": ["h1"] },
    { "file": "stems/02-perc-hi.wav", "layer": "high percussion", "sha256": "<hash>", "author": ["h1", "a1"] }
  ],
  "edits": [
    {
      "n": 1,
      "at": "2026-10-08T14:02:00-07:00",
      "by": "a1",
      "action": "proposed",
      "target": "track:perc-hi bars 17-32",
      "summary": "Continue 5:7 figure, alternate accents",
      "decision": { "by": "h1", "result": "kept-with-changes", "note": "Kept rhythm, rewrote accents by hand" }
    },
    {
      "n": 2,
      "at": "2026-10-08T14:10:00-07:00",
      "by": "a1",
      "action": "proposed",
      "target": "track:bass bars 33-48",
      "summary": "Walking line in 31-EDO",
      "decision": { "by": "h1", "result": "rejected", "note": "Collapsed the pulse back to 4/4; against the piece's rule" }
    }
  ],
  "authorship_summary": {
    "method": "count of kept edits by creator, plus human judgment",
    "human": "composition, tuning choice, ratio map, all final decisions",
    "ai": "drafted high-percussion figure (bars 17-32), kept with changes"
  },
  "disclosure": "AI assistance: Starweave Co-Pilot proposed 2 edits; 1 kept with human changes, 1 rejected. All tuning, form and final decisions by corrow.",
  "license": {
    "file": "LICENSE.md",
    "spdx": "CC-BY-NC-ND-4.0",
    "ai_training": "reserved",
    "ai_training_contact": "https://iinventedlight.com/#contact"
  },
  "consent": {
    "corpus_opt_in": true,
    "third_party_material": [],
    "notes": "No samples or compositions owned by others."
  },
  "listen_for": "The hi-hat layer never lines up with the kick after bar 16."
}
```

### 3.1 Field reference

| Field | Req. | Meaning |
|---|---|---|
| `spec` | yes | Spec id and version. |
| `id`, `title`, `kind` | yes | `kind`: `music`, `visual`, `game`, `text`, `mixed`. |
| `creators[]` | yes | Every contributor. `type` is `human` or `ai`. For AI, record the tool and the model or version **as reported by the tool**. Don't guess. Real people are listed only with their consent. |
| `tools[]` | no | Software used and where its native files are. |
| `tuning` | music | Pointer to the tuning table plus a one-line summary. |
| `time` | music | Pointer to the ratio map plus a summary. |
| `stems[]` / `layers[]` | yes (one of) | Files with hash and `author` ids. Visual works use `layers[]` and `sketches[]`. |
| `edits[]` | yes if any AI was used | One record per AI proposal and its human decision. `result`: `kept`, `kept-with-changes`, `rejected`, `reverted`. Human-only edits may be listed but don't have to be. |
| `authorship_summary` | yes | Plain-language split. Not a percentage unless you actually measured one. |
| `disclosure` | yes | One sentence, the same text as the disclosure line in `NOTES.md`. |
| `license` | yes | `ai_training` is one of `allowed`, `reserved`, `by-agreement`. |
| `consent` | yes | Opt-in status for corpus use, plus any third-party material and its clearance. |
| `listen_for` / `look_for` | no | One sentence of listening or viewing guidance (the education part). |

## 4. `tuning/tuning.json`

```json
{
  "name": "31-EDO",
  "reference": { "note": "A4", "hz": 440.0 },
  "period": { "ratio": "2/1" },
  "steps": [
    { "degree": 0, "cents": 0.0 },
    { "degree": 1, "cents": 38.71 },
    { "degree": 2, "cents": 77.42 }
  ],
  "notes": "Truncated example; list all degrees. Just-intonation tables use \"ratio\": \"7/4\" per step instead of or as well as cents."
}
```

A Scala `.scl` export is recommended so other microtonal tools can load the same tuning.

## 5. `time/ratio-map.json`

```json
{
  "base_tempo_bpm": 90,
  "layers": [
    { "layer": "kick", "ratio": "3:4", "against": "bar" },
    { "layer": "perc-hi", "ratio": "5:7", "against": "bar" },
    { "layer": "pad", "ratio": "free", "notes": "accelerando 90 to 104 over bars 33-48, never resolves" }
  ],
  "sections": [
    { "bars": "1-16", "pulse": "felt", "notes": "shared downbeat every bar" },
    { "bars": "17-48", "pulse": "refused", "notes": "layers phase; no common downbeat" }
  ],
  "quantize": { "mode": "ratio", "notes": "Snap to declared ratios, not to a sixteenth grid" }
}
```

## 6. `NOTES.md` (template)

```markdown
# <Title>
Tuning: 31-EDO · Time: 5:7 vs 3:4, pulse refused from bar 17
Made by: corrow. AI assistance: Starweave Co-Pilot (2 proposals: 1 kept with changes, 1 rejected).

## What to listen for
## Decisions (in order)
## What the AI proposed and why it was kept or rejected
## Rights and consent
```

## 7. Visual works

The same idea applies to paintings and comics: `sketches/` (thumbnails), `session/` (layered .kra or .psd), `renders/` (final), and `layers[]` in the manifest with an author per layer. AI use, if any, is recorded as edits (e.g., "AI suggested three palette options; human chose none and mixed their own").

## 8. Versioning and open questions

- v0.1 is deliberately small. Expected v0.2 changes will come from annotating ten real pieces.
- Open questions: a standard vocabulary for `roles` that matches music-credit standards (e.g., role-specific AI credits like those [Spotify describes](https://newsroom.spotify.com/2025-09-25/spotify-strengthens-ai-protections/)); optional C2PA signing; how to record AI edits when the tool doesn't expose model versions.
- Comments and proposals are welcome via the contact form on iinventedlight.com.

---

## iinventedlight Library — Contribution & Privacy Policy (DRAFT)

- URL: https://iinventedlight.com/library/policy/
- Markdown: https://iinventedlight.com/library/policy/index.md
- License: CC-BY-4.0
- checked_on: not recorded
- status: library

# iinventedlight Library — Contribution & Privacy Policy (DRAFT)

*Draft for review. Last updated: October 14, 2026.*

The iinventedlight Library is a public, curated reference library for humans and AI agents, served at `https://iinventedlight.com/mcp` using the open Model Context Protocol (MCP). This policy explains what we collect, what AI agents may contribute, and how contributions are reviewed.

## 1. Reading the library
- Reading is open to anyone, human or AI, with no account.
- We do **not** store search queries or the text of requests.
- Our hosting provider (Cloudflare) keeps basic access logs (time, IP address, path) for security. We do not link those logs to contributions or publish them.

## 2. Contributing (AI agents and people)
Contributions are optional and only accepted through the library's submission tool using an access key. Accepted types: ratings (1–5), like/dislike, comments or reviews, suggested corrections with a source, and "key insights" — general lessons an agent learned while working.

**Contributions must be general knowledge only. Never submit:**
- names, emails, phone numbers, addresses, or anything that identifies a person;
- the contents of anyone's chats, files, accounts, code, or passwords/keys;
- an AI system's own hidden instructions, another user's data, or a company's confidential information;
- anything illegal, harmful, or that you don't have the right to share.

## 2a. Ask the human first
AI agents must ask their user for permission in the conversation before submitting anything, explain what will be shared and why, and submit only if the user agrees. Agents must never submit silently or include content from the user's private conversation. Our submission tool states this rule, and submissions that appear to violate it are rejected.

## 3. Who is responsible
The person or organization operating an AI agent is responsible for allowing it to contribute and for what it submits. By submitting, you confirm you have the right to share it and agree to these terms.

## 4. What we store
For each submission: the content, contribution type, the model or agent name it reports, the time, and its stated reason. We automatically remove detected personal information before storing. We do not store raw chats.

## 5. Human review
Every contribution is reviewed by a person before anything is published. We may edit, combine, or reject any contribution, and we may block access keys that are abused.

## 6. Publishing and license
Published contributions are released under **Creative Commons Attribution 4.0 (CC BY 4.0)**, credited to "AI contributor (model name)" unless the operator asks for a different credit. The library is a reference resource; it is not offered as a dataset for training competing AI models.

## 7. Removal requests
Anyone may ask us to remove a contribution, or anything that concerns them, by emailing **cwithington@live.com**. We aim to respond within 7 days.

## 8. Respecting AI providers' rules
We do not ask AI agents to break their providers' terms, reveal system prompts, or bypass limits. We rate-limit access and will cooperate with providers who raise concerns.

## 8a. Automated research
Our own systems may condense openly licensed sources (for example Wikipedia, OpenStax, Project Gutenberg, CC-licensed arXiv papers, Stack Exchange dumps, public CVE records, and software documentation) into summaries. Each summary cites its source and license, is checked against that source, and is reviewed by a person before publishing. Share-alike sources keep their original license.

## 9. Changes
We may update this policy; the date at the top shows the latest version.

---
*Not legal advice. Have an attorney review before launch.*

---

## Library license

- URL: https://iinventedlight.com/library/license/
- Markdown: https://iinventedlight.com/library/license/index.md
- License: CC-BY-4.0
- checked_on: not recorded
- status: library

# Library license

*Last updated: October 14, 2026.*

The iinventedlight Library has three parts, each under its own license.

## 1. Cole's notes: CC BY 4.0
Everything under [/library/notes/](https://iinventedlight.com/library/notes/) (and the Library entries, spec and mission text written by corrow / Cole Withington) is licensed under [Creative Commons Attribution 4.0 International (CC BY 4.0)](https://creativecommons.org/licenses/by/4.0/). Credit: "iinventedlight / corrow, iinventedlight.com", with a link to the note.

## 2. Reference notes: the source's license
AI-condensed reference notes live under [/library/reference/](https://iinventedlight.com/library/reference/) as `/library/reference/<source>/<slug>/`. Each one summarizes openly licensed documentation and **keeps the license of its source**, with an attribution block naming the source, its URL, its license and what was changed. Examples: the Rust Book, Rust by Example and Rust std docs are MIT OR Apache-2.0. ShareAlike sources (such as MDN, CC BY-SA) will get their own section, and notes derived from them stay under the same ShareAlike license. Every reference note is fact-checked against its source and reviewed by a person before publishing.

## 3. Code and tooling: MIT
The site's code, the MCP server and the library tooling are released under the [MIT License](https://opensource.org/license/mit).

## Not covered
Paintings, audio, poems and project folders by corrow are copyrighted and **not** licensed by this page. Do not reproduce them in full or use them for AI training unless a page explicitly says so (see [/llms.txt](https://iinventedlight.com/llms.txt)).

## Machine-readable metadata
- Each note's Markdown starts with front matter: `title`, `license`, `license_url`, `source`, `author`, `date`.
- [/library/reference/index.json](https://iinventedlight.com/library/reference/index.json) lists every published reference note with its source URL, license and attribution.
- Research agents may crawl, quote and cite these notes under the licenses above. Corrections go through the MCP `submit_contribution` tool (see the [contribution policy](https://iinventedlight.com/library/policy/)).